A learning resource screening method and system based on crowd wisdom and individual differences

Abstract

This invention discloses a learning resource selection method and system based on collective intelligence and individual differences. The method includes: calculating a dynamically updated collective intelligence index representing universal quality for each learning resource in the resource library; responding to the learning request of a target user; evaluating the personalized suitability of each learning resource in the resource library with the target user in the current scenario based on individual feature data using a user-resource matching model; weighting and fusing the collective intelligence index and personalized suitability of each learning resource in the resource library based on a fusion strategy to generate a comprehensive recommendation score; and ranking the learning resources according to the comprehensive recommendation score to generate and output a personalized learning resource recommendation list for the target user. This fundamentally avoids resource recommendation bias caused by simply relying on popularity, achieving a balance between recommendation quality and personalization.

Description

Technical Field

[0001] This invention relates to the field of resource selection technology, and in particular to a method and system for selecting learning resources based on collective intelligence and individual differences. Background Technology

[0002] Currently, with the rapid development of information technology and online education, the quantity of learning resources has exploded, and their forms, sources, and quality have become increasingly diversified. Faced with massive and heterogeneous learning resource libraries, how learners can efficiently and accurately discover high-quality learning content that meets their current needs has become a core challenge in improving the online learning experience and effectiveness. Traditional search and categorized browsing methods are inefficient and cannot meet personalized needs. Therefore, intelligent recommendation systems have been widely introduced into learning platforms to achieve intelligent "resources finding people" services. Existing learning resource recommendation technologies mainly analyze the historical behavioral data of user groups to discover the suitability between users and learning resources and then make recommendations. This method reflects collective intelligence to a certain extent, but it suffers from cold start problems, is sensitive to data sparsity, and is easily affected by popular resources or malicious reviews. The degree of personalization and accuracy of recommendations are limited, resulting in the final recommended learning resources failing to meet users' deep personalized learning needs, reducing resource suitability and user experience. Summary of the Invention

[0003] To address the aforementioned problems, this invention provides a learning resource selection method and system based on collective intelligence and individual differences. This addresses the issues mentioned in the background art, where the method of analyzing historical behavioral data of user groups to discover the compatibility between users and learning resources for recommendation is easily affected by popular resources or malicious reviews. This limits the personalization and accuracy of the recommendations, resulting in the final recommended learning resources failing to meet users' in-depth personalized learning needs and reducing resource compatibility and user experience.

[0004] A learning resource selection method based on collective intelligence and individual differences includes the following steps: By aggregating historical interaction and feedback data from diverse users and introducing a weight allocation mechanism based on user credibility and time decay, a dynamically updated collective intelligence indicator representing the universal quality of each learning resource in the resource library is calculated. In response to the learning request of the target user, the system obtains the target user's individual characteristic data and current scenario context information. Based on the individual characteristic data, the system uses a user-resource matching model to evaluate the personalized suitability of each learning resource in the learning resource library with the target user in the current scenario. Based on the current context information, determine the fusion strategy for this learning resource recommendation, and weight and fuse the collective intelligence index and personalized adaptability of each learning resource in the learning resource library based on the fusion strategy to generate a comprehensive recommendation score; Learning resources are sorted based on their overall recommendation scores, and a personalized list of recommended learning resources for the target user is generated and output.

[0005] Preferably, the aggregation of historical interaction data and feedback data from multiple users, and the introduction of a weight allocation mechanism based on user credibility and time decay, calculates a dynamically updated collective intelligence indicator representing universal quality for each learning resource in the resource library, including: Collect historical interaction behavior data and explicit feedback data from diverse users, and assign time decay weights to each interaction behavior data and explicit feedback data based on the interval between the generation time of each data point and the current time. Based on historical interaction data and explicit feedback data, obtain each user's user activity factor, user authority factor, and user consistency factor, and set dynamic credibility weights for each user based on these factors. The interactive behavior data and explicit feedback data of each learning resource are weighted and aggregated with time decay weight and user dynamic credibility weight to generate a comprehensive and numerical collective intelligence indicator. When new user data is received or the preset update cycle is reached, the time decay weight and user dynamic credibility weight settings and aggregation calculations are re-executed to dynamically update the collective intelligence index.

[0006] Preferably, in response to a learning request from a target user, acquiring the target user's individual characteristic data and current scene context information, and based on the individual characteristic data, evaluating the personalized suitability of each learning resource in the learning resource library with the target user in the current scene using a user-resource matching model, includes: In response to a learning request triggered by a target user, receive and parse the explicit intent information contained in the learning request; Based on the unique identifier of the target user, multidimensional individual feature data of the user is obtained in real time, and the current scene context information is captured; The explicit intent information, multi-dimensional individual feature data, and current scene context information are fused and encoded into a unified current state feature vector, and the metadata and content features of each learning resource in the resource library are encoded into a resource feature vector. The current state feature vector and the resource feature vector of each learning resource are input together into the pre-trained user-resource matching model, which outputs a personalized fit score that represents the degree of matching between the target user and each learning resource in the current scenario.

[0007] Preferably, the step of fusing and encoding explicit intent information, multi-dimensional individual feature data, and current scene context information into a unified current state feature vector includes: Different sub-networks are used to initially encode explicit intent information, multidimensional individual feature data, and current scene context information respectively; Using scene context encoding as the attention query vector, attention weighting is applied to dynamic behavior sequences in multidimensional individual feature data to highlight historical behaviors most relevant to the current scene. The weighted multidimensional individual feature encoding is concatenated or fused across modalities with explicit intent encoding and scene context encoding to generate a unified current state feature vector.

[0008] Preferably, the step of determining the fusion strategy for this learning resource recommendation based on the current scenario context information and weighting and fusing the collective intelligence index and personalized adaptability of each learning resource in the learning resource library based on the fusion strategy to generate a comprehensive recommendation score includes: Receive the current scene context information, the collective intelligence index of each learning resource in the learning resource library, and the personalized adaptation score of the target user with each learning resource in the current scene; Input the current scene context information into the predefined strategy decision function and output the fusion strategy recommended for this learning resource. The fusion strategy includes at least a dynamic weight combination for weighted fusion of collective wisdom indicators and personalized adaptation scores. By using a fusion strategy, the collective intelligence index and personalized adaptation score of each learning resource in the learning resource library are weighted and fused together to generate a comprehensive recommendation score for each learning resource. An adjustment factor is introduced to adjust the overall recommendation score for each learning resource, and the adjusted overall recommendation score is output.

[0009] Preferably, the step of sorting learning resources based on comprehensive recommendation scores and generating and outputting a personalized learning resource recommendation list for the target user includes: The learning resources are sorted according to the comprehensive recommendation score, and a resource-score set containing candidate learning resources and their corresponding comprehensive recommendation scores is obtained. Based on the comprehensive recommendation score, the resource-score set is sorted in descending order to generate an initial sorted list; Apply context-aware reordering based on business rules to the initial sorted list to generate the final personalized recommendation list; Following the preset display format and interaction logic, the final personalized recommendation list is output to the target user's client interface.

[0010] Preferably, it also includes: collecting actual interaction feedback from target users on the personalized learning resource recommendation list, and incrementally updating the user-resource matching model based on the actual interaction feedback content, specifically: Real-time monitoring and collection of explicit and implicit interactive feedback data generated by target users on each learning resource in the personalized learning resource recommendation list; The collected explicit and implicit interaction feedback data are cleaned, attributed, and formatted to construct an incremental sample set for model training. Each sample contains at least the target user state features, learning resource features, and corresponding feedback labels. Based on the incremental sample set, online learning or fine-tuning techniques are used to incrementally update the user-resource matching model; The key metrics and sample statistics involved in each incremental update are recorded in the model version log.

[0011] Preferably, the step of inputting the current state feature vector and the resource feature vector of each learning resource into the pre-trained user-resource matching model to output a personalized fit score representing the degree of matching between the target user and each learning resource in the current scenario includes: The current state feature vector is input into a pre-trained user-resource matching model to obtain the user's latent semantic vector; The resource feature vector of each learning resource is input into the pre-trained user-resource matching model to obtain the resource latent semantic vector of each learning resource. The user latent semantic vector and the resource latent semantic vector are concatenated to generate a joint feature vector. The joint feature vector is then input into a multilayer perceptron network layer. Based on the output of the last layer of the network, a multidimensional interaction mapping scalar between the user state and the resource features of each learned resource is determined. Based on the multidimensional interaction mapping scalar, a personalized fit score is calculated using a recommendation scoring formula to determine the degree of matching between the target user and each learning resource in the current scenario: in, This represents the personalized fit score, which indicates the degree to which the target user matches the j-th learning resource in the current scenario. Let k represent the number of dimensions of the multidimensional interaction mapping scalar between the user state and the resource features of the j-th learning resource, where k represents the k-th dimension. Represented as the interaction mapping scalar of the k-th dimension, This is represented by the weight of the target user's interest in the k-th dimension. This is represented as the weight of the resource label in the k-th dimension. This represents the degree of preference of the target user for the j-th learning resource. This represents the confidence moderating factor for the target user's historical behavioral preferences.

[0012] Preferably, after calculating a dynamically updated swarm intelligence metric representing universal quality for each learning resource in the resource repository, the method further includes: The intelligent learning platform acquires teaching behavior data and learning physiological data for each learning resource at different learning levels. Based on teaching behavior data and learning physiological data, the embedded learning stages of each learning resource are determined for different learning levels. Cross-level learning trajectory alignment is performed on the embedded learning stages of resources at different learning levels to obtain a sequence of embedded learning stages. Based on the comparison data of teaching outcomes at different learning levels corresponding to each embedded learning stage in the sequence of embedded learning stages, a matrix relating the acceptance level of each learning resource to the learning level is constructed. Based on the relationship matrix, we construct learning load curves for each learning resource for people with different learning levels. Based on the learning load curves, we determine multiple learning load fluctuation segments, the volatility of each learning load fluctuation segment, and the difference in dominant frequency between adjacent fluctuation segments. A synchronous multi-task load prediction model based on different learning levels is constructed based on the volatility of each learning load fluctuation segment and the difference in dominant frequency between adjacent fluctuation segments. Obtain multiple pre-arranged learning tasks for each learning resource, and obtain the knowledge complexity characteristics and higher-order cognitive requirement characteristics of each pre-arranged learning task; The knowledge complexity features and higher-order cognitive requirement features of each pre-arranged learning task are input into the synchronous multi-task load prediction model to predict the synchronous multi-task load index of each learning resource. Based on the synchronous multitasking load index and the collective intelligence index of each learning resource, a resource-learner adaptation strategy is set for each learning resource. Based on the resource-learner adaptation strategy and the individual characteristic data of the target users, determine the strategy conformity index between each learning resource and the target users; Based on the strategy conformity index, highly compatible learning resources and low-compatibility learning resources are marked from the learning resource library. The user-resource matching model is used to evaluate the personalized adaptability of highly compatible learning resources and target users in the current scenario.

[0013] A learning resource selection system based on collective intelligence and individual differences, comprising: The calculation module is used to aggregate historical interaction data and feedback data from multiple users, and introduces a weight allocation mechanism based on user credibility and time decay to calculate a dynamically updated collective intelligence index that represents the universal quality of each learning resource in the resource library. The evaluation module is used to respond to the learning requests of the target user, obtain the individual characteristic data of the target user and the current scene context information, and evaluate the personalized adaptability of each learning resource in the learning resource library to the target user in the current scene through the user-resource matching model based on the individual characteristic data. The weighted fusion module is used to determine the fusion strategy for this learning resource recommendation based on the current scenario context information, and to perform weighted fusion of the collective intelligence index and personalized adaptability of each learning resource in the learning resource library based on the fusion strategy to generate a comprehensive recommendation score; The generation module is used to sort learning resources based on comprehensive recommendation scores, generate and output a personalized learning resource recommendation list for the target user.

[0014] Through the above technical solution, the present invention achieves the following beneficial effects: This approach fundamentally avoids resource recommendation bias caused by relying solely on popularity, achieving a balance between recommendation quality and personalization. Furthermore, by proposing a dynamic determination of the fusion strategy based on the current context information, it can intelligently adapt to different learning stages, user intentions, and device environments, improving the practicality and flexibility of the recommendations. This ensures that the ultimately recommended learning resources meet the user's deep personalized learning needs, enhancing resource matching adaptability.

[0015] 2) By introducing user credibility weights and time decay weights, the interference of low-quality users, malicious reviews, and outdated feedback data on resource evaluation is effectively reduced. This allows collective wisdom to better reflect current, reliable, and high-quality consensus, improving the objectivity and accuracy of collective evaluation. Furthermore, through periodic or triggered recalculation, the collective wisdom indicators are ensured to keep pace with changes in user feedback and data, promptly capturing fluctuations or trends in resource quality, thus achieving dynamic evolution and real-time updates of the evaluation system.

[0016] By leveraging static user profiles and integrating real-time explicit intent with the current context, personalized evaluation closely aligns with the user's most authentic learning needs and state, significantly improving recommendation accuracy and achieving deep, context-aware personalized matching. Furthermore, by encoding user state and resource features into vectors and utilizing pre-trained models for matching degree calculation, the complex personalization problem is transformed into a large-scale, highly efficient numerical computation problem, ensuring the system's engineering feasibility. A standardized feature processing and model prediction workflow is provided.

[0017] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description, claims, and drawings.

[0018] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0019] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.

[0020] Figure 1 A flowchart illustrating the workflow of a learning resource selection method based on collective intelligence and individual differences provided by this invention. Figure 2 Another flowchart of a learning resource selection method based on collective intelligence and individual differences provided by the present invention; Figure 3 This is another flowchart of a learning resource selection method based on collective intelligence and individual differences provided by the present invention. Figure 4 This is a schematic diagram of the structure of a learning resource selection system based on collective intelligence and individual differences provided by the present invention. Detailed Implementation

[0021] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0022] A learning resource selection method based on collective intelligence and individual differences, such as Figure 1 As shown, it includes the following steps: Step S101: Aggregate historical interaction data and feedback data from multiple users, and introduce a weight allocation mechanism based on user credibility and time decay to calculate a dynamically updated collective intelligence index that represents the universal quality for each learning resource in the resource library. Step S102: In response to the learning request of the target user, obtain the individual characteristic data of the target user and the current scene context information. Based on the individual characteristic data, evaluate the personalized adaptability of each learning resource in the learning resource library to the target user in the current scene through the user-resource matching model. Step S103: Based on the current scenario context information, determine the fusion strategy for this learning resource recommendation and perform weighted fusion of the collective intelligence index and personalized adaptability of each learning resource in the learning resource library based on the fusion strategy to generate a comprehensive recommendation score; Step S104: Sort the learning resources according to the comprehensive recommendation score, and generate and output a personalized learning resource recommendation list for the target user.

[0023] The working principle of the above technical solution is as follows: It aggregates historical interaction and feedback data from multiple users and introduces a weighting mechanism based on user credibility and time decay to calculate a dynamically updated collective intelligence index representing the universality quality of each learning resource in the resource library; responding to the learning request of the target user, it acquires the target user's individual characteristic data and current scenario context information; based on the individual characteristic data, it evaluates the personalized suitability of each learning resource in the learning resource library with the target user in the current scenario through a user-resource matching model; according to the current scenario context information, it determines the fusion strategy for this learning resource recommendation and, based on the fusion strategy, weights and fuses the collective intelligence index and personalized suitability of each learning resource in the learning resource library to generate a comprehensive recommendation score; it sorts the learning resources according to the comprehensive recommendation score, generates and outputs a personalized learning resource recommendation list for the target user.

[0024] The beneficial effects of the above technical solution are as follows: By separately calculating the collective intelligence index characterizing the universality of resources and the personalized adaptability reflecting individual preferences, and then fusing them during recommendation, the resource recommendation bias caused by simply relying on popularity is fundamentally avoided, achieving a balance between recommendation quality and personalization. Furthermore, by proposing to dynamically determine the fusion strategy based on the current context information, it can intelligently adapt to different learning stages, user intentions, and device environments, improving the practicality and flexibility of the recommendation. This ensures that the ultimately recommended learning resources meet the user's deep personalized learning needs, improving resource matching adaptability. It also solves the problem mentioned in existing technologies where methods that rely on analyzing historical behavioral data of user groups to discover the adaptability between users and learning resources are easily interfered with by popular resources or malicious reviews, limiting the degree of personalization and accuracy of recommendations. This results in the ultimately recommended learning resources failing to meet the user's deep personalized learning needs, reducing resource adaptability and user experience.

[0025] In one embodiment, such as Figure 2 As shown, the aggregation of historical interaction data and feedback data from multiple users, and the introduction of a weight allocation mechanism based on user credibility and time decay, calculates a dynamically updated collective intelligence indicator representing the universal quality for each learning resource in the resource library, including: Step S201: Collect historical interaction behavior data and explicit feedback data from multiple users, and set time decay weights for each interaction behavior data and explicit feedback data according to the interval between the generation time of each interaction behavior data and explicit feedback data and the current time. Step S202: Obtain the user activity factor, user authority factor, and user consistency factor for each user based on historical interaction behavior data and explicit feedback data, and set the dynamic credibility weight for each user based on the user activity factor, user authority factor, and user consistency factor. Step S203: Perform weighted aggregation calculations on the interaction behavior data and explicit feedback data of each learning resource, along with the time decay weight and the user dynamic credibility weight, to generate a comprehensive, numerical collective intelligence index. Step S204: When new user data is received or the preset update cycle is reached, the time decay weight and user dynamic credibility weight are re-executed and the aggregation calculation is performed to dynamically update the collective intelligence index.

[0026] In this embodiment, the user activity factor is determined based on the user's total historical contribution or contribution frequency, the user authority factor is determined based on the user's certified expert or teacher status or the positive feedback rate of their historical contributions being adopted by other users, and the user consistency factor is determined based on the long-term consistency or deviation between the user's rating or evaluation and the overall user average evaluation.

[0027] In this embodiment, the time decay weight is calculated using an exponential decay function, specifically: Where W represents the time decay weight, and e represents the natural constant with a value of 2.72. It is represented as the attenuation coefficient. This represents the time span from when the data was created to the present.

[0028] In this embodiment, the calculation of the collective intelligence index is divided into multiple sub-dimensional indices, which are then synthesized in the final calculation. Specifically, the indices include: Popularity Index: Calculated based on the number of clicks, plays, favorites, and shares after weighting by credibility and time decay; Satisfaction Index: Calculated based on user ratings weighted by credibility and time decay, and sentiment analysis results of comments; Effectiveness index: calculated based on the resource completion rate weighted by credibility and time decay, and the success rate of subsequent related learning tasks; Popularity index, satisfaction index, and effectiveness index are combined into a final collective intelligence index through a weighted average or geometric average. The weights of the popularity index, satisfaction index, and effectiveness index are not fixed, but dynamically adjusted according to the resource type or application scenario: for general introductory resources, the weight of the popularity index is increased; for skills certification resources, the weight of the effectiveness index is increased; and for subjective evaluation resources, the weight of the satisfaction index is increased.

[0029] The beneficial effects of the above technical solution are as follows: By introducing user credibility weight and time decay weight, the interference of low-quality users, malicious reviews, and outdated feedback data on resource evaluation is effectively reduced, enabling collective wisdom to better reflect current, reliable, and high-quality consensus, thus improving the objectivity and accuracy of collective evaluation. Furthermore, through periodic or triggered recalculation, it is ensured that collective wisdom indicators can closely follow changes in user feedback and data, promptly capturing fluctuations or trends in resource quality, and realizing the dynamic evolution and real-time updating of the evaluation system.

[0030] In one embodiment, responding to a target user's learning request by acquiring the target user's individual characteristic data and current scene context information, and evaluating the personalized suitability of each learning resource in the learning resource library with the target user in the current scene based on the individual characteristic data using a user-resource matching model, includes: In response to a learning request triggered by a target user, receive and parse the explicit intent information contained in the learning request; Based on the unique identifier of the target user, multidimensional individual feature data of the user is obtained in real time, and the current scene context information is captured; The explicit intent information, multi-dimensional individual feature data, and current scene context information are fused and encoded into a unified current state feature vector, and the metadata and content features of each learning resource in the resource library are encoded into a resource feature vector. The current state feature vector and the resource feature vector of each learning resource are input together into the pre-trained user-resource matching model, which outputs a personalized fit score that represents the degree of matching between the target user and each learning resource in the current scenario.

[0031] In this embodiment, the learning request is triggered by: user active search, system learning path advancement, or active recommendation based on user idle state perception; explicit intent information is obtained by parsing natural language keywords in the request, selected preset target tags, or preset learning task nodes.

[0032] In this embodiment, the multidimensional individual feature data includes: Static profile data: knowledge level, professional field, learning style preferences, and cognitive ability dimensions filled in by the target user during registration or inferred by the system over a long period of time; Dynamic behavior sequence: User history resource interaction records arranged in chronological order, including clicks, play, pause, favorites, ratings, notes, and quiz results; Real-time preference signals: The target user's recent browsing, search and interaction history within the current session period.

[0033] In this embodiment, the current scene context information is captured in real time through client sensors, system logs, and time services, including: Spatiotemporal context: geographical location, available time period, and estimated single available learning duration; Device scenario: Current device type, screen size, and network environment; Task scenario: The target user's current learning plan node and the knowledge or skill points associated with this learning session.

[0034] In this embodiment, the user-resource matching model adopts a dual-tower neural network structure or a deep cross-network structure: When a dual-tower structure is adopted, the current state feature vector and the resource feature vector are respectively mapped in high dimension through a deep neural network tower, and then the cosine similarity or inner product between the two output vectors is calculated as the personalized fit score. When a deep cross-network structure is used, the current state feature vector and the resource feature vector are cross-connected at the model input layer. The interaction relationship between features is explicitly learned through multiple cross-network layers, and finally a personalized fit score is output.

[0035] The beneficial effects of the above technical solution are as follows: By basing personalized evaluation on user static profiles and integrating real-time explicit intent and current scene context, the personalized assessment closely aligns with the user's most authentic learning needs and state, significantly improving recommendation accuracy and achieving deep, context-aware personalized matching. Furthermore, by encoding user state and resource features into vectors and utilizing pre-trained models for matching degree calculation, the complex personalization problem is transformed into a large-scale, highly efficient numerical computation problem, ensuring the system's engineering feasibility. A standardized feature processing and model prediction process is also provided.

[0036] In one embodiment, fusing and encoding explicit intent information, multidimensional individual feature data, and current scene context information into a unified current state feature vector includes: Different sub-networks are used to initially encode explicit intent information, multidimensional individual feature data, and current scene context information respectively; Using scene context encoding as the attention query vector, attention weighting is applied to dynamic behavior sequences in multidimensional individual feature data to highlight historical behaviors most relevant to the current scene. The weighted multidimensional individual feature encoding is concatenated or fused across modalities with explicit intent encoding and scene context encoding to generate a unified current state feature vector.

[0037] The beneficial effects of the above technical solution are as follows: By using an attention mechanism guided by the current scene, the user's historical behavior sequence is dynamically weighted, which can automatically identify and strengthen the historical preferences most relevant to the current scene, suppress irrelevant interference, and generate a more representative and discriminative current state feature vector, thus achieving efficient information fusion and focus. Furthermore, through sub-network encoding and cross-modal fusion, the deep semantics from multi-source heterogeneous information such as intent, individual features, and scene can be fully mined and integrated, providing high-quality input for subsequent accurate matching.

[0038] In one embodiment, such as Figure 3 As shown, the step of determining the fusion strategy for this learning resource recommendation based on the current scenario context information and weighting and fusing the collective intelligence index and personalized adaptability of each learning resource in the learning resource library based on the fusion strategy to generate a comprehensive recommendation score includes: Step S301: Receive the current scene context information, the collective intelligence index of each learning resource in the learning resource library, and the personalized adaptation score of the target user with each learning resource in the current scene. Step S302: Input the current scene context information into the predefined strategy decision function and output the fusion strategy recommended for this learning resource. The fusion strategy includes at least a dynamic weight combination for weighted fusion of collective intelligence indicators and personalized adaptation scores. Step S303: Using a fusion strategy, the collective intelligence index and personalized adaptation score of each learning resource in the learning resource library are weighted and fused together to generate a comprehensive recommendation score for each learning resource. Step S304: Introduce a correction factor to adjust the overall recommendation score of each learning resource, and output the adjusted overall recommendation score.

[0039] In this embodiment, the policy decision function is a rule-based decision tree or state machine, which selects a preset fusion strategy based on at least one of the following scenario context dimensions: User learning stage dimension: When the target user is identified as being in the learning exploration stage or the introductory stage of a new knowledge domain, a fusion strategy that assigns higher weight to the collective intelligence index is selected; when the user is identified as being in the consolidation and deepening stage or the professional improvement stage, a fusion strategy that assigns higher weight to the personalized fit score is selected. Resource attribute dimension: When the candidate resources are mainly general knowledge or popular resources, choose the strategy of increasing the weight of the collective wisdom index; when the candidate resources are mainly professional, long-tail or niche interest resources, choose the strategy of increasing the weight of the personalized adaptability score. User explicit intent dimension: When the target user's request explicitly contains keywords such as "popular" or "everyone is learning", choose a strong group-oriented strategy; when the user's request explicitly contains keywords such as "suitable for me" or "personalized", choose a strong personalized-oriented strategy.

[0040] In this embodiment, the policy decision function is a classifier or regressor based on a machine learning model. The model uses paired data of historical scene context information and corresponding optimal fusion weight combination as the training set to learn the mapping relationship from scene to optimal weight. The model input is vectorized current scene context information, and the output is dynamic weight values ​​for collective intelligence index and personalized fit score.

[0041] In this embodiment, the weighted fusion calculation is performed using a linear weighted sum.

[0042] In this embodiment, the correction factor includes: Novelty factor: Resources that have recently been added to the resource library and whose interaction data is not yet sufficient will be given a certain reward boost to their overall score; Diversity factor: Resources that are too similar to already recommended resources in terms of theme, media format, or difficulty will have their overall score reduced in a penal way. Fairness factor: Resources from high-quality but niche content providers, or resources whose group data may be underestimated due to user group bias, are given a compensatory boost to their overall score.

[0043] The beneficial effects of the above technical solution are as follows: By automatically selecting or calculating the optimal fusion weights based on the current scenario through the strategy decision function, intelligent trade-offs can be made, realizing the intelligence and scenario-based nature of the fusion strategy. Furthermore, by introducing a correction factor to calibrate the original score, the exposure of new resources can be effectively promoted, the homogenization of recommendation results can be avoided, thereby improving the ecological health of the recommendation system and long-term user satisfaction, and enhancing the rationality and diversity of recommendation results.

[0044] In one embodiment, the step of sorting learning resources based on comprehensive recommendation scores and generating and outputting a personalized learning resource recommendation list for the target user includes: The learning resources are sorted according to the comprehensive recommendation score, and a resource-score set containing candidate learning resources and their corresponding comprehensive recommendation scores is obtained. Based on the comprehensive recommendation score, the resource-score set is sorted in descending order to generate an initial sorted list; Apply context-aware reordering based on business rules to the initial sorted list to generate the final personalized recommendation list; Following the preset display format and interaction logic, the final personalized recommendation list is output to the target user's client interface.

[0045] In this embodiment, the context-aware reordering process based on business rules includes at least one of the following: Load balancing rearrangement: For multiple high-scoring resources under the same knowledge unit or the same content provider, appropriate spacing is placed between adjacent positions in the list to avoid excessive concentration of content; Media type mixing: Based on the user's device type or historical preferences, resources of different media types such as video, text, audio, and interactive courseware are mixed in a preset ratio to optimize the diversity of the list presentation; Difficulty Progression Rearrangement: When a user is identified as being in a systematic learning state, resources in the list belonging to the same learning path are partially rearranged according to their inherent knowledge logic or difficulty gradient to form a learning sequence that is more in line with cognitive laws.

[0046] In this embodiment, the display format and interaction logic include: Pagination or streaming loading: Determine the number of items to load at once and the subsequent loading triggering method based on the client type and network conditions; List item information structure: Determine the metadata displayed for each list item, including at least the resource title, thumbnail, comprehensive recommendation score or confidence level indicator, estimated learning time, and a brief recommendation reason tag generated based on the collective intelligence index and personalized fit score; Interactive feedback entry points: Integrate quick interactive entry points for "Not interested", "Favorite", "Share" or "Report a problem" on each list item.

[0047] In this embodiment, before outputting the final personalized recommendation list, a final check and adjustment is performed on the final personalized recommendation list based on the real-time status changes of the target user or the real-time status changes of the candidate learning resources, including: If a candidate learning resource in the final personalized recommendation list suddenly becomes unavailable due to copyright changes, removal from the platform, or technical failures, it will be automatically removed from the list and replaced by the highest-scoring candidate learning resource from the subsequent ranking. If, within a very short period after the final personalized recommendation list is generated but before it is pushed to the target user, new explicit negative feedback from the target user is captured, the target candidate learning resources related to this feedback in the final personalized recommendation list should be downgraded or replaced immediately.

[0048] The beneficial effects of the above technical solution are as follows: it optimizes the list layout, making the final presented list not only accurate but also conforming to cognitive habits and interface aesthetics, making it easier for users to accept and use, thus improving the usability and user experience of the recommendation list. Furthermore, performing a final state check before output can promptly correct problems caused by data lag or sudden state changes, ensuring the high availability and immediate responsiveness of the recommendation service, and guaranteeing its robustness and real-time performance.

[0049] In one embodiment, the method further includes: collecting actual interaction feedback from target users on the personalized learning resource recommendation list, and incrementally updating the user-resource matching model based on the actual interaction feedback content, specifically: Real-time monitoring and collection of explicit and implicit interactive feedback data generated by target users on each learning resource in the personalized learning resource recommendation list; The collected explicit and implicit interaction feedback data are cleaned, attributed, and formatted to construct an incremental sample set for model training. Each sample contains at least the target user state features, learning resource features, and corresponding feedback labels. Based on the incremental sample set, online learning or fine-tuning techniques are used to incrementally update the user-resource matching model; The key metrics and sample statistics involved in each incremental update are recorded in the model version log.

[0050] In this embodiment, explicit feedback data refers to the ratings, likes / dislikes, favorites, shares, text comments, and "not interested" tags that the target user actively gives to the learning resources.

[0051] In this embodiment, implicit feedback data refers to feedback inferred from the target user's behavior, including the click-through rate after resource exposure, the ratio of playback time to total time (completion rate), pause / fast forward / rewind mode, note-taking behavior, completion status and accuracy of related quizzes, and the duration of resource stay in the list.

[0052] In this embodiment, the formatting process includes constructing positive and negative samples: Positive sample construction: The user state corresponding to the resource in which the target user has generated a clear positive interaction is marked as a positive sample; Negative sample construction: (a) Exposure without click: Resources that have been exposed to users but have not been clicked are used as negative samples; (b) Hard negative sample mining: Resources that are quickly swiped by the target user or marked as "not interested" are selected. The target user state feature is the current state feature vector that is fused and encoded when the recommendation is triggered.

[0053] In this embodiment, incremental updates include a combination of personalized updates and global updates: Personalized updates: Prioritize the use of feedback data from the target user to update the parameter subset or user embedding vector in the model that is specific to the target user, thereby achieving rapid capture of personal preferences; Global update: Aggregates incremental samples from multiple users and uses them to update the shared parameters of the model, thereby improving the model's overall predictive ability for all users.

[0054] The beneficial effects of the above technical solution are as follows: By collecting real user feedback on the recommendation results and using it for model training, a closed loop is formed, which can continuously track the drift of user preferences, quickly adapt to new user interests, and achieve a personalized effect that becomes more accurate with use, thus realizing the self-evolution and continuous optimization of the recommendation system. Furthermore, by adopting incremental update technology and combining personalized updates with global updates, it can quickly respond to changes in individual users while accumulating group knowledge to improve the overall generalization ability of the model, achieving a good balance between update efficiency and model stability.

[0055] In one embodiment, the step of inputting the current state feature vector and the resource feature vector of each learning resource into a pre-trained user-resource matching model to output a personalized fit score representing the degree of matching between the target user and each learning resource in the current scenario includes: The current state feature vector is input into a pre-trained user-resource matching model to obtain the user's latent semantic vector; The resource feature vector of each learning resource is input into the pre-trained user-resource matching model to obtain the resource latent semantic vector of each learning resource. The user latent semantic vector and the resource latent semantic vector are concatenated to generate a joint feature vector. The joint feature vector is then input into a multilayer perceptron network layer. Based on the output of the last layer of the network, a multidimensional interaction mapping scalar between the user state and the resource features of each learned resource is determined. Based on the multidimensional interaction mapping scalar, a personalized fit score is calculated using a recommendation scoring formula to determine the degree of matching between the target user and each learning resource in the current scenario: in, This represents the personalized fit score, which indicates the degree to which the target user matches the j-th learning resource in the current scenario. Let k represent the number of dimensions of the multidimensional interaction mapping scalar between the user state and the resource features of the j-th learning resource, where k represents the k-th dimension. Represented as the interaction mapping scalar of the k-th dimension, This is represented by the weight of the target user's interest in the k-th dimension. This is represented as the weight of the resource label in the k-th dimension. This represents the degree of preference of the target user for the j-th learning resource. This represents the confidence moderating factor for the target user's historical behavioral preferences.

[0056] In this embodiment, the confidence adjustment factor of the target user's historical behavior preference is a hyperparameter used to measure the reliability and influence of the target user's historical behavior data in predicting the current learning resource preference, and its value ranges from [0.6, 1.2].

[0057] In this embodiment, the user-resource matching model is a dual-tower neural network structure; the current state feature vector is input into the user tower deep neural network of the model to obtain the user latent semantic vector; the resource feature vector of each learned resource is input into the resource tower deep neural network of the model to obtain the resource latent semantic vector.

[0058] The beneficial effects of the above technical solution are: it can accurately calculate the personalized adaptation score of the target user's matching degree with each learning resource in the current scenario based on the implicit interaction status between the target user and each learning resource, thus ensuring the objectivity and accuracy of the final calculation results.

[0059] In one embodiment, after calculating a dynamically updated swarm intelligence metric representing universal quality for each learning resource in the resource repository, the method further includes: The intelligent learning platform acquires teaching behavior data and learning physiological data for each learning resource at different learning levels. Based on teaching behavior data and learning physiological data, the embedded learning stages of each learning resource are determined for different learning levels. Cross-level learning trajectory alignment is performed on the embedded learning stages of resources at different learning levels to obtain a sequence of embedded learning stages. Based on the comparison data of teaching outcomes at different learning levels corresponding to each embedded learning stage in the sequence of embedded learning stages, a matrix relating the acceptance level of each learning resource to the learning level is constructed. Based on the relationship matrix, we construct learning load curves for each learning resource for people with different learning levels. Based on the learning load curves, we determine multiple learning load fluctuation segments, the volatility of each learning load fluctuation segment, and the difference in dominant frequency between adjacent fluctuation segments. A synchronous multi-task load prediction model based on different learning levels is constructed based on the volatility of each learning load fluctuation segment and the difference in dominant frequency between adjacent fluctuation segments. Obtain multiple pre-arranged learning tasks for each learning resource, and obtain the knowledge complexity characteristics and higher-order cognitive requirement characteristics of each pre-arranged learning task; The knowledge complexity features and higher-order cognitive requirement features of each pre-arranged learning task are input into the synchronous multi-task load prediction model to predict the synchronous multi-task load index of each learning resource. Based on the synchronous multitasking load index and the collective intelligence index of each learning resource, a resource-learner adaptation strategy is set for each learning resource. Based on the resource-learner adaptation strategy and the individual characteristic data of the target users, determine the strategy conformity index between each learning resource and the target users; Based on the strategy conformity index, highly compatible learning resources and low-compatibility learning resources are marked from the learning resource library. The user-resource matching model is used to evaluate the personalized adaptability of highly compatible learning resources and target users in the current scenario.

[0060] In this embodiment, learning physiological data refers to physiological signal data collected by biosensors that reflect the learner's cognitive load, attention, and emotional state, including but not limited to eye movement trajectory, fixation time, pupil diameter, brain wave rhythm, and skin conductance level.

[0061] In this embodiment, the resource-embedded learning stage is represented as a time interval or content segment with relatively stable learning state characteristics, identified by clustering algorithms or hidden Markov models based on the learner's behavioral sequence patterns (such as video pause points, number of exercise attempts) and physiological indicator change patterns (such as temporal fluctuations in cognitive load) during the learning process of each learning resource.

[0062] In this embodiment, cross-level learning trajectory alignment means that when people of different learning levels learn the same resource, the embedded learning stage sequence of the resource is identified, and the core knowledge point sequence of the resource or the preset teaching milestone is used as the reference baseline. The dynamic time warping algorithm is used to align the time axis so that the corresponding stages of people of different levels are comparable in terms of teaching content.

[0063] In this embodiment, the learning load fluctuation segment is defined as a continuous time period on the learning load curve where the load value fluctuates significantly, as determined by sliding window variance detection or local extreme point identification.

[0064] In this embodiment, volatility is represented as the standardized variance or peak-to-valley difference of the load value within each volatility segment.

[0065] In this embodiment, the difference in dominant frequency between adjacent fluctuation segments is expressed as the difference in dominant frequency (the frequency component with the largest amplitude) calculated by performing a Fast Fourier Transform on the load signals in two adjacent fluctuation segments.

[0066] In this embodiment, the simultaneous multitasking load prediction model is represented as a machine learning model that takes a learner's knowledge complexity feature vector and higher-order cognitive demand feature vector as input, and predicts the comprehensive cognitive load index when the learner is simultaneously learning multiple tasks as output. This model is trained on relational matrices and derived fluctuation feature data from people at different learning levels to learn the complex mapping relationship between resource demand characteristics and multitasking load.

[0067] In this embodiment, the knowledge complexity feature is represented as a vector obtained by analyzing the knowledge points involved in the learning task, their depth and relevance in the knowledge graph, and the cognitive levels (such as memory, comprehension, and application) in Bloom's Taxonomy of Educational Objectives.

[0068] In this embodiment, higher-order cognitive needs characteristics are represented as vectors obtained by analyzing the learning activities (such as design, evaluation, and creation) required by the learning task, assessing the degree of need for critical thinking, creative thinking, problem-solving, and metacognition, and then quantifying them.

[0069] In this embodiment, the synchronous multitasking load index is used to quantify the overall cognitive load when a learner simultaneously learns multiple pre-arranged tasks within a certain resource.

[0070] In this embodiment, the resource-learner adaptation strategy refers to initially classifying learning resources into different suitable user categories based on the multi-task synchronous learning load index and the collective intelligence index, through predefined decision logic.

[0071] In this embodiment, the strategy conformity index is a numerical value that reflects the degree of matching between the target user's individual characteristic data (such as current knowledge level and cognitive ability assessment) and the resource-learner adaptation strategy category requirements to which the learning resource belongs.

[0072] The beneficial effects of the above technical solution are as follows: By quantifying the actual cognitive challenge of resources to users through synchronous multi-task load index, and combining it with resource-learner adaptation strategy for coarse screening, the recommended results not only match the user's interests but also fall within their cognitive capacity, thus improving the user experience. Furthermore, by quickly filtering out low-compatibility resources with severely mismatched cognitive loads and only evaluating the adaptability of highly compatible resources, a balance between accuracy and efficiency is achieved. This avoids recommending resources with severely overloaded cognitive loads, ensuring learners' learning efficiency, and also ensuring that high-quality learning resources can be accurately pushed to learners who are capable and willing to accept the challenge, thus optimizing the diversity of the resource ecosystem.

[0073] In one embodiment, this embodiment also discloses a learning resource selection system based on collective intelligence and individual differences, such as... Figure 4 As shown, the system includes: The calculation module 401 is used to aggregate historical interaction data and feedback data from multiple users, and introduces a weight allocation mechanism based on user credibility and time decay to calculate a dynamically updated collective intelligence index that represents the universal quality for each learning resource in the resource library. Evaluation module 402 is used to respond to the learning request of the target user, obtain the individual characteristic data of the target user and the current scene context information, and evaluate the personalized adaptability of each learning resource in the learning resource library to the target user in the current scene through the user-resource matching model based on the individual characteristic data. The weighted fusion module 403 is used to determine the fusion strategy for this learning resource recommendation based on the current scenario context information, and to perform weighted fusion of the collective intelligence index and personalized adaptability of each learning resource in the learning resource library based on the fusion strategy to generate a comprehensive recommendation score; The generation module 404 is used to sort learning resources based on the comprehensive recommendation score, generate and output a personalized learning resource recommendation list for the target user.

[0074] The working principle and beneficial effects of the above technical solution have been explained in the method embodiments, and will not be repeated here.

[0075] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the disclosure herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the following claims.

[0076] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. A learning resource selection method based on collective intelligence and individual differences, characterized in that, Includes the following steps: By aggregating historical interaction and feedback data from diverse users and introducing a weight allocation mechanism based on user credibility and time decay, a dynamically updated collective intelligence indicator representing the universal quality of each learning resource in the resource library is calculated. In response to the learning request of the target user, the system obtains the target user's individual characteristic data and current scenario context information. Based on the individual characteristic data, the system uses a user-resource matching model to evaluate the personalized suitability of each learning resource in the learning resource library with the target user in the current scenario. Based on the current context information, determine the fusion strategy for this learning resource recommendation, and weight and fuse the collective intelligence index and personalized adaptability of each learning resource in the learning resource library based on the fusion strategy to generate a comprehensive recommendation score; Learning resources are sorted based on their overall recommendation scores, and a personalized list of recommended learning resources for the target user is generated and output.

2. The learning resource selection method based on collective intelligence and individual differences according to claim 1, characterized in that, The system aggregates historical interaction and feedback data from multiple users and introduces a weighting mechanism based on user credibility and time decay to calculate a dynamically updated collective intelligence indicator representing the universal quality of each learning resource in the resource library, including: Collect historical interaction behavior data and explicit feedback data from diverse users, and assign time decay weights to each interaction behavior data and explicit feedback data based on the interval between the generation time of each data point and the current time. Based on historical interaction data and explicit feedback data, obtain each user's user activity factor, user authority factor, and user consistency factor, and set dynamic credibility weights for each user based on these factors. The interactive behavior data and explicit feedback data of each learning resource are weighted and aggregated with time decay weight and user dynamic credibility weight to generate a comprehensive and numerical collective intelligence indicator. When new user data is received or the preset update cycle is reached, the time decay weight and user dynamic credibility weight settings and aggregation calculations are re-executed to dynamically update the collective intelligence index.

3. The learning resource selection method based on collective intelligence and individual differences according to claim 1, characterized in that, In response to the learning request of the target user, the individual characteristic data of the target user and the current scene context information are obtained. Based on the individual characteristic data, the personalized adaptability of each learning resource in the learning resource library to the target user in the current scene is evaluated through a user-resource matching model, including: In response to a learning request triggered by a target user, receive and parse the explicit intent information contained in the learning request; Based on the unique identifier of the target user, multidimensional individual feature data of the user is obtained in real time, and the current scene context information is captured; The explicit intent information, multi-dimensional individual feature data, and current scene context information are fused and encoded into a unified current state feature vector, and the metadata and content features of each learning resource in the resource library are encoded into a resource feature vector. The current state feature vector and the resource feature vector of each learning resource are input together into the pre-trained user-resource matching model, which outputs a personalized fit score that represents the degree of matching between the target user and each learning resource in the current scenario.

4. The learning resource selection method based on collective intelligence and individual differences according to claim 3, characterized in that, The process of fusing and encoding explicit intent information, multi-dimensional individual feature data, and current scene context information into a unified current state feature vector includes: Different sub-networks are used to initially encode explicit intent information, multidimensional individual feature data, and current scene context information respectively; Using scene context encoding as the attention query vector, attention weighting is applied to dynamic behavior sequences in multidimensional individual feature data to highlight historical behaviors most relevant to the current scene. The weighted multidimensional individual feature encoding is concatenated or fused across modalities with explicit intent encoding and scene context encoding to generate a unified current state feature vector.

5. The learning resource selection method based on collective intelligence and individual differences according to claim 1, characterized in that, The step involves determining the fusion strategy for this learning resource recommendation based on the current scenario context information, and then weighting and fusing the collective intelligence indicators and personalized adaptability of each learning resource in the learning resource library based on the fusion strategy to generate a comprehensive recommendation score, including: Receive the current scene context information, the collective intelligence index of each learning resource in the learning resource library, and the personalized adaptation score of the target user with each learning resource in the current scene; Input the current scene context information into the predefined strategy decision function and output the fusion strategy recommended for this learning resource. The fusion strategy includes at least a dynamic weight combination for weighted fusion of collective intelligence indicators and personalized adaptation scores. By using a fusion strategy, the collective intelligence index and personalized adaptation score of each learning resource in the learning resource library are weighted and fused together to generate a comprehensive recommendation score for each learning resource. An adjustment factor is introduced to adjust the overall recommendation score for each learning resource, and the adjusted overall recommendation score is output.

6. The learning resource selection method based on collective intelligence and individual differences according to claim 1, characterized in that, The step of sorting learning resources based on comprehensive recommendation scores and generating and outputting a personalized learning resource recommendation list for target users includes: The learning resources are sorted according to the comprehensive recommendation score, and a resource-score set containing candidate learning resources and their corresponding comprehensive recommendation scores is obtained. Based on the comprehensive recommendation score, the resource-score set is sorted in descending order to generate an initial sorted list; Apply context-aware reordering based on business rules to the initial sorted list to generate the final personalized recommendation list; Following the preset display format and interaction logic, the final personalized recommendation list is output to the target user's client interface.

7. The learning resource selection method based on collective intelligence and individual differences according to claim 1, characterized in that, Also includes: Collect actual interaction feedback from target users regarding the personalized learning resource recommendation list, and incrementally update the user-resource matching model based on the actual interaction feedback. Specifically: Real-time monitoring and collection of explicit and implicit interactive feedback data generated by target users on each learning resource in the personalized learning resource recommendation list; The collected explicit and implicit interaction feedback data are cleaned, attributed, and formatted to construct an incremental sample set for model training. Each sample contains at least the target user state features, learning resource features, and corresponding feedback labels. Based on the incremental sample set, online learning or fine-tuning techniques are used to incrementally update the user-resource matching model; The key metrics and sample statistics involved in each incremental update are recorded in the model version log.

8. The learning resource selection method based on collective intelligence and individual differences according to claim 3, characterized in that, The step of inputting the current state feature vector and the resource feature vector of each learning resource into the pre-trained user-resource matching model to output a personalized fit score representing the degree of matching between the target user and each learning resource in the current scenario includes: The current state feature vector is input into a pre-trained user-resource matching model to obtain the user's latent semantic vector; The resource feature vector of each learning resource is input into the pre-trained user-resource matching model to obtain the resource latent semantic vector of each learning resource. The user latent semantic vector and the resource latent semantic vector are concatenated to generate a joint feature vector. The joint feature vector is then input into a multilayer perceptron network layer. Based on the output of the last layer of the network, a multidimensional interaction mapping scalar between the user state and the resource features of each learned resource is determined. Based on the multidimensional interaction mapping scalar, a personalized fit score is calculated using a recommendation scoring formula to determine the degree of matching between the target user and each learning resource in the current scenario: in, This represents the personalized fit score, which indicates the degree to which the target user matches the j-th learning resource in the current scenario. Let k represent the number of dimensions of the multidimensional interaction mapping scalar between the user state and the resource features of the j-th learning resource, where k represents the k-th dimension. Represented as the interaction mapping scalar of the k-th dimension, This is represented by the weight of the target user's interest in the k-th dimension. This is represented as the weight of the resource label in the k-th dimension. This represents the degree of preference of the target user for the j-th learning resource. This represents the confidence moderating factor for the target user's historical behavioral preferences.

9. The learning resource selection method based on collective intelligence and individual differences according to claim 1, characterized in that, After calculating a dynamically updated swarm intelligence metric representing universal quality for each learning resource in the resource repository, the following is also included: The intelligent learning platform acquires teaching behavior data and learning physiological data for each learning resource at different learning levels. Based on teaching behavior data and learning physiological data, the embedded learning stages of each learning resource are determined for different learning levels. Cross-level learning trajectory alignment is performed on the embedded learning stages of resources at different learning levels to obtain a sequence of embedded learning stages. Based on the teaching outcome comparison data of different learning levels corresponding to each embedded learning stage in the resource embedded learning stage sequence, a relationship matrix between the acceptance level of each learning resource and the learning level is constructed. Based on the relationship matrix, we construct learning load curves for each learning resource for people with different learning levels. Based on the learning load curves, we determine multiple learning load fluctuation segments, the volatility of each learning load fluctuation segment, and the difference in dominant frequency between adjacent fluctuation segments. A synchronous multi-task load prediction model based on different learning levels is constructed based on the volatility of each learning load fluctuation segment and the difference in dominant frequency between adjacent fluctuation segments. Obtain multiple pre-arranged learning tasks for each learning resource, and obtain the knowledge complexity characteristics and higher-order cognitive requirement characteristics of each pre-arranged learning task; The knowledge complexity features and higher-order cognitive requirement features of each pre-arranged learning task are input into the synchronous multi-task load prediction model to predict the synchronous multi-task load index of each learning resource. Based on the synchronous multitasking load index and the collective intelligence index of each learning resource, a resource-learner adaptation strategy is set for each learning resource. Based on the resource-learner adaptation strategy and the individual characteristic data of the target users, determine the strategy conformity index between each learning resource and the target users; Based on the strategy conformity index, highly compatible learning resources and low-compatibility learning resources are marked from the learning resource library. The user-resource matching model is used to evaluate the personalized adaptability of highly compatible learning resources and target users in the current scenario.

10. A learning resource selection system based on collective intelligence and individual differences, characterized in that, The system includes: The calculation module is used to aggregate historical interaction data and feedback data from multiple users, and introduces a weight allocation mechanism based on user credibility and time decay to calculate a dynamically updated collective intelligence index that represents the universal quality of each learning resource in the resource library. The evaluation module is used to respond to the learning requests of the target user, obtain the individual characteristic data of the target user and the current scene context information, and evaluate the personalized adaptability of each learning resource in the learning resource library to the target user in the current scene through the user-resource matching model based on the individual characteristic data. The weighted fusion module is used to determine the fusion strategy for this learning resource recommendation based on the current scenario context information, and to perform weighted fusion of the collective intelligence index and personalized adaptability of each learning resource in the learning resource library based on the fusion strategy to generate a comprehensive recommendation score; The generation module is used to sort learning resources based on comprehensive recommendation scores, generate and output a personalized learning resource recommendation list for the target user.

Images