Control method and control device of learning machine and learning machine

Abstract

The invention provides a learning machine control method and device and a learning machine, and relates to the technical field of learning machines, and the learning machine control method comprises the steps: obtaining the identification information of a user, and determining a learning habit corresponding to the user; obtaining a historical learning feature record corresponding to the knowledge source description in response to a typing operation for the knowledge source description; generating an optimal learning mode corresponding to the user according to the learning habit and the historical learning feature record; generating a corresponding learning task sequence according to the optimal learning mode; displaying the learning task sequence, and obtaining execution feedback of the user on the learning task sequence; adjusting the priority and the content difficulty of the learning task sequence according to the execution feedback, optimizing the learning path of the user, and updating the historical learning feature record. The method can solve the problem that a personalized learning path is rigid in the prior art, can dynamically adjust the learning task sequence according to the user characteristics and real-time feedback, and improves the learning efficiency and experience.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of learning machines, and in particular to a learning machine control method, a learning machine control device and a learning machine. BACKGROUND

[0002] The control logic of existing learning machines excessively relies on pre-set static content architecture, such as fixed chapter order or coarse-grained knowledge classification system, and general recommendation rules, lacks fine-grained modeling of individual user characteristics, and cannot dynamically decide and adapt to the best learning method that truly fits the user and the current knowledge content. SUMMARY

[0003] The present application aims to improve learning efficiency and experience.

[0004] To achieve the above-mentioned purpose, the present application provides a learning machine control method, which comprises: obtaining the identification information of the user and determining the learning habit corresponding to the user; in response to a key-in operation for a knowledge source description, obtaining a historical learning feature record corresponding to the knowledge source description; generating a best learning method corresponding to the user according to the learning habit and the historical learning feature record; generating a corresponding learning task sequence according to the best learning method; displaying the learning task sequence and obtaining the execution feedback of the user on the learning task sequence; adjusting the priority and content difficulty of the learning task sequence according to the execution feedback, optimizing the learning path of the user, and updating the historical learning feature record.

[0005] Optionally, the generating of the best learning method corresponding to the user according to the learning habit and the historical learning feature record comprises: extracting a multi-dimensional behavior feature vector of the user when learning different knowledge types in the historical learning feature record, the multi-dimensional behavior feature vector comprising knowledge point switching frequency, single knowledge point residence time length distribution, and exercise accuracy rate change pattern over time; obtaining user long-term preference information in the learning habit, the user long-term preference information comprising selection tendency for graphical, sequential or aural content; generating a comprehensive learning feature portrait according to the multi-dimensional behavior feature vector and the user long-term preference information; based on the comprehensive learning feature portrait, matching a preset learning method template library to determine the best learning method that is most suitable for the current knowledge source description of the user.

[0006] Optionally, the generating a comprehensive learning feature portrait according to the multi-dimensional behavior feature vector and the long-term preference information of the user comprises: extracting the multi-dimensional behavior feature vector representing the user's learning stability, concentration decay rate, and knowledge internalization efficiency; quantifying the selection tendency data in the long-term preference information of the user to generate a confidence score for graphical, sequential, and auditory content; matrix fusion of the learning stability, concentration decay rate, and knowledge internalization efficiency with the corresponding confidence score to generate a comprehensive learning feature portrait; the matching of the comprehensive learning feature portrait with the preset learning method template library to determine the best learning method that best fits the user's current knowledge source description, comprising: performing in-depth semantic analysis on the knowledge source description to generate a corresponding lightweight knowledge graph, the lightweight knowledge graph including core concept nodes, logical relationships between concepts, and attribute labels of knowledge domains; cross-dimension alignment of the comprehensive learning feature portrait and the lightweight knowledge graph to determine the adaptation score between the user's cognitive mode and the knowledge structure characteristics; According to the adaptation score, the optimal matching item is retrieved from the preset learning method template library to determine the best learning method.

[0007] Optionally, the generating a corresponding learning task sequence according to the best learning method comprises: calling the knowledge structuring conversion rule corresponding to the best learning method; According to the knowledge structuring conversion rule, the knowledge content corresponding to the knowledge source description is converted to generate a knowledge node graph with logical levels; According to the user's single learning duration and rhythm preference in the learning habit, the knowledge node graph is divided into multiple learning task blocks, each of which has independent logical coherence; Generating an executable learning task and a feedback node corresponding to each learning task block, the feedback node including an embedded multiple-choice question, a knowledge repetition trigger point, or a scenario simulation operation; The executable learning tasks are sorted by priority to generate the learning task sequence.

[0008] Optionally, the acquiring the historical learning feature record corresponding to the knowledge source description in response to the typing operation on the knowledge source description comprises: In response to the knowledge source description input by the user terminal, the key theme words and the main body classification of the knowledge source description are analyzed to determine the subject field and the knowledge dimension to which the knowledge source belongs; retrieve historical learning feature records associated with the knowledge source from a local cache based on the subject field and the knowledge dimension; if no matching record is retrieved, establish a communication connection with a cloud server; retrieve historical learning feature records matching the knowledge source description from the cloud server.

[0009] Optionally, the adjusting the priority and content difficulty of the learning task sequence based on the execution feedback, optimizing the learning path of the user, and updating the historical learning feature records comprise: analyzing the execution feedback to determine knowledge weak points, understanding speed abnormal points, and interest decay points exhibited by the user in the learning task sequence; determining content difficulty adjustment amounts for corresponding task nodes in the learning task sequence based on the knowledge weak points and the understanding speed abnormal points, the content difficulty adjustment amounts including adding pre-laid information, changing the explanation perspective, or decomposing complex steps; re-evaluating the priority of tasks in the learning task sequence that have not been executed based on the interest decay points, and generating a corresponding task priority adjustment strategy; reconstructing the learning task sequence based on the content difficulty adjustment amounts and the task priority adjustment strategy to optimize the learning path of the user; updating the optimized learning path to the historical learning feature records.

[0010] Optionally, the learning machine comprises a display component; the displaying the learning task sequence and obtaining the execution feedback of the user on the learning task sequence comprises: controlling the display component to render and display the learning task sequence in the form of a process chart or a list; in response to a selection operation on any task node in the learning task sequence, retrieving the learning content corresponding to the task node; in response to an interaction operation on the learning content, collecting time series data of the user in content browsing, pausing, backtracking, accelerating, and answering behavior; generating the execution feedback based on the time series data, the execution feedback including task completion degree, interaction frequency, and abnormal interval of stay time.

[0011] Optionally, the obtaining the identification information of the user and determining the learning habits of the user comprise: obtaining the unique identification information of the user through a user login credential and accessing a cloud user archive; retrieve a long-period history learning summary report of the user from the cloud user repository, the long-period history learning summary report including click rate, completion rate and active search preference of the user on different media form learning resources; construct a user multi-dimensional learning preference portrait based on the long-period history learning summary report, identify the tendency of the user on text, video or interactive content, and generate the learning habit.

[0012] In addition, to achieve the above-mentioned purpose, the application also provides a control device, which comprises a memory, a processor and a control program of a learning machine stored on the memory and executable on the processor, and the control program of the learning machine is configured to implement the control method of the learning machine as described above.

[0013] In addition, to achieve the above-mentioned purpose, the application also provides a learning machine, which comprises: a display component for presenting a learning task sequence and an interactive interface; and a control device as described above, which is electrically connected with the display component; The control device is used to obtain the identification information of the user, determine the learning habit corresponding to the user, and in response to the typing operation on the knowledge source description, obtain the historical learning feature record corresponding to the knowledge source description, and then generate the optimal learning mode corresponding to the user according to the learning habit and the historical learning feature record, and generate the corresponding learning task sequence according to the optimal learning mode, display the learning task sequence, obtain the execution feedback of the user on the learning task sequence, adjust the priority and content difficulty of the learning task sequence according to the execution feedback, optimize the learning path of the user, and update the historical learning feature record.

[0014] The embodiment of the application obtains the identification information of the user, determines the learning habit corresponding to the user, and in response to the typing operation on the knowledge source description, obtains the historical learning feature record corresponding to the knowledge source description, and then generates the optimal learning mode corresponding to the user according to the learning habit and the historical learning feature record, generates the corresponding learning task sequence according to the optimal learning mode, displays the learning task sequence, obtains the execution feedback of the user on the learning task sequence, and finally adjusts the priority and content difficulty of the learning task sequence according to the execution feedback, optimizes the learning path of the user, and updates the historical learning feature record. In this way, the optimal learning mode is generated by obtaining the user habit and historical record, and the priority and difficulty of the task sequence are dynamically adjusted according to the real-time execution feedback, which can solve the problem of rigid personalized learning path in the prior art, dynamically adjust the learning task sequence according to the user features and real-time feedback, and improve the learning efficiency and experience. BRIEF DESCRIPTION OF DRAWINGS

[0015] The accompanying drawings, which are incorporated herein and constitute part of the specification, illustrate embodiments consistent with the application and, together with the description, further serve to explain the principles of the application.

[0016] Figure 1 Flowchart of a control method of a learning machine according to an embodiment of the application; Figure 2 Flowchart of a control method of a learning machine according to another embodiment of the application; Figure 3 Flowchart of a control method of a learning machine according to yet another embodiment of the application; Figure 4 Flowchart of a control method of a learning machine according to still another embodiment of the application; Figure 5 Flowchart of a control method of a learning machine according to yet another embodiment of the application; Figure 6 Flowchart of a control method of a learning machine according to another embodiment of the application; Figure 7 Flowchart of a control method of a learning machine according to yet another embodiment of the application; Figure 8 Flowchart of a control method of a learning machine according to still another embodiment of the application.

[0017] The implementation, functional features and advantages of the application will be described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION

[0018] With the deep integration of educational informatization and artificial intelligence technology, intelligent learning devices are increasingly popular in personalized learning assistance. Current learning machines generally have basic functional modules, including a knowledge content repository, an interactive question and answer engine, and a learning progress tracking system. During use, users can input keywords or select preset knowledge topics through a keyboard or touch interface, and the system retrieves and presents standardized learning materials accordingly. In addition, the device continuously records user learning behavior data, such as correct answer rate and error patterns, knowledge point browsing time distribution, and task completion sequence, and generates basic learning reports or recommends supplementary content of similar difficulty based on these data.

[0019] However, such systems have significant defects in achieving dynamic generation and closed-loop optimization of deep personalized learning paths. The control logic of existing learning machines relies too much on pre-set static content organization frameworks, such as fixed chapter order or coarse-grained knowledge classification systems, and generalized recommendation rules, lacking fine-grained modeling of individual user characteristics. Although the system can identify user identity and accumulate historical data, it fails to effectively integrate the user's long-term learning habit characteristics, historical learning behavior patterns in specific knowledge fields, and real-time interaction feedback during the learning process, resulting in the inability to dynamically decide and adapt to the best learning method that truly fits the user and the current knowledge content. For example, facing the same knowledge point, some users may understand better through tree diagram logical deduction, some users rely on classification memory method for induction, and some users may have better results in the context of story immersion, but existing technologies cannot accurately match such learning methods according to multi-dimensional characteristics. In addition, the system has difficulty in continuously dynamically adjusting the priority of the learning task sequence, the content difficulty gradient, and the knowledge presentation path based on real-time feedback (such as fluctuations in understanding speed, shifts in points of interest, or exposure of weak knowledge links) when the user performs the learning task. This static and rigid learning path design causes users to frequently encounter inadequate adaptability during the learning process, resulting in low learning efficiency, excessive cognitive load, and fragmented learning experience, which seriously hinders the realization of personalized learning goals.

[0020] The main solution of the embodiment of the present application is: by acquiring the identification information of the user, determining the learning habit corresponding to the user, and in response to the key-in operation for the knowledge source description, acquiring the historical learning feature record corresponding to the knowledge source description, and according to the learning habit and the historical learning feature record, generating the best learning method corresponding to the user, and then generating the corresponding learning task sequence according to the best learning method, displaying the learning task sequence, and acquiring the execution feedback of the user to the learning task sequence, finally adjusting the priority and content difficulty of the learning task sequence according to the execution feedback, optimizing the learning path of the user, and updating the historical learning feature record.

[0021] In this embodiment, for ease of description, the following describes the control device as the execution subject.

[0022] The present application provides a solution that generates the best learning method by acquiring user habits and historical records, and dynamically adjusts the priority and difficulty of the task sequence according to real-time execution feedback, which can solve the problem of rigid personalized learning path in the prior art, dynamically adjust the learning task sequence according to user characteristics and real-time feedback, and improve the advantages of learning efficiency and experience.

[0023] To this end, the application provides a control method of a learning machine; it can be understood that the learning machine is provided with a control device for storing and executing the following method, and the control device can be implemented by a main controller, such as an MCU (Microcontroller Unit), a DSP (Digital Signal Process), an FPGA (Field Programmable Gate Array), an SOC (System On Chip), etc.

[0024] The existing intelligent learning machine generates a personalized learning path by relying on a pre-set static content organization mode and a general recommendation rule, and fails to deeply integrate long-term learning habits of a user, historical learning features in a specific knowledge field, and real-time learning feedback, so that it is difficult to dynamically decide and adapt to an optimal learning mode. In addition, the existing system is also difficult to continuously and closed-loop optimize and adjust a priority of a learning task sequence, content difficulty and a presentation path based on real-time feedback of a user during execution of the learning task, so that the learning path has insufficient adaptability, and learning efficiency and experience promotion encounter a bottleneck.

[0025] Reference Figure 1 In an embodiment of the application, the control method of the learning machine comprises steps S100-S600, wherein: S100, acquiring identification information of a user to determine learning habits of the user; S200, acquiring historical learning feature records corresponding to a knowledge source description in response to a typing operation for the knowledge source description; S300, generating an optimal learning mode corresponding to the user according to the learning habits and the historical learning feature records; S400, generating a corresponding learning task sequence according to the optimal learning mode; S500, displaying the learning task sequence and acquiring execution feedback of the user for the learning task sequence; S600, adjusting a priority and content difficulty of the learning task sequence according to the execution feedback, optimizing a learning path of the user, and updating the historical learning feature records.

[0026] Among them, learning habits refer to the relatively stable learning preferences, cognitive patterns, and behavioral rules formed by users in the long-term learning process. Learning habits can include user's tendency to different media forms (such as text, image, video, audio) content, preference for learning rhythm (such as single learning duration, rest interval), and adaptability to knowledge presentation mode (such as logical deduction, induction summary, situation simulation). Learning habits are an important basis for the control device to plan personalized learning paths. Knowledge source description refers to the text or symbolic information provided by the user to the learning machine through keyboard operation or other input methods to indicate the knowledge content or topic they want to learn. It can be a keyword, a phrase, a question, or even the name of a knowledge field. The knowledge source description is used to guide the control device to retrieve and organize related learning resources.

[0027] Among them, historical learning feature records refer to the long-term accumulation of behavioral data and performance data about users learning specific knowledge points or knowledge fields by the control device. Historical learning feature records can include user's learning duration on different knowledge types, practice accuracy, knowledge point switching frequency, and acceptance of specific learning methods, etc. The historical learning feature records reflect the strengths and weaknesses of users in past learning, and are an important reference for the control device to assess the user's current learning state and predict future learning effectiveness. The best learning method refers to the knowledge presentation and interaction mode that can maximize user learning efficiency and effectiveness, which is intelligently recommended by the control device according to the user's learning habits, historical learning feature records, and the characteristics of the current knowledge source. The best learning method can include but not limited to tree diagram, classification memory method, situational story method, question and answer interaction method, etc. The best learning method aims to provide the most suitable cognitive path for users.

[0028] Among them, the learning task sequence refers to a series of learning activities arranged in a certain logical order and priority for users to complete step by step after structuring and modularizing knowledge content according to the determined best learning method. Each task node can be the learning of a knowledge point, the completion of an exercise, or the repetition of a concept. The learning task sequence aims to guide users to systematically master knowledge. The execution feedback refers to the user behavior data and learning result data collected by the control device during the execution of the learning task sequence. It can include task completion, answer accuracy, interaction frequency, content browsing duration, pause and backtrack behavior, and user's active expression of interest or confusion, etc. The execution feedback is the key information for the system to dynamically adjust the learning path and optimize the learning experience.

[0029] In this embodiment, the identification information of the user is first acquired, and the learning habit of the user is determined. In one implementation, the control device can prompt the user to manually input his / her tendency of learning content form (e.g. preferring to read text, watch video or do interactive exercises), and his / her desired single learning duration and learning pace. For example, the user can check “I prefer to learn through a combination of text and images, and each time the learning duration is not more than 30 minutes”. In this way, the control device can preliminarily understand the learning preference of the user.

[0030] Secondly, when the user inputs a keyword or phrase (e.g. “Newton’s first law”) as the knowledge source description in the learning machine interface, the control device can directly match the keyword and search the local stored database for whether there is a historical learning record completely consistent with the keyword. If there is, the record is directly extracted.

[0031] Then, the optimal learning mode corresponding to the user is generated according to the learning habit and the historical learning feature record. For example, the control device can judge based on the learning preference (e.g. preferring visual learning) manually input by the user and his / her mastering situation (e.g. still having errors after multiple exercises) of a specific knowledge point (e.g. physical concept) shown in the historical record through a preset simple rule. If the user prefers visual learning and has a weak point in the physical concept, the control device can recommend “tree diagram” as the optimal learning mode to help the user sort out the logical relationship between concepts.

[0032] Subsequently, the control device can structure the knowledge content corresponding to the knowledge source description according to the selected optimal learning mode (e.g. “tree diagram”), for example, decompose it into several independent knowledge points. Then, these knowledge points are arranged in a preset order (e.g. from basic to advanced) to form a learning task list. For example, if the optimal learning mode is “tree diagram”, the task sequence can include “introduction of concept A”, “relationship between concept A and concept B”, “instance of concept B”, etc.

[0033] Further, the control device can present the generated learning task sequence in the form of a text list on the display interface. When the user clicks on a task in the list, the control device will load the corresponding learning content. After the user completes the task, the control device will record whether the task is marked as “completed” or not, and take it as part of the execution feedback. For example, after the user completes a reading task, he / she can manually click the “completed” button.

[0034] Finally, if the control device receives feedback from the user that a certain task has been completed, the priority of the task will be lowered, and the priorities of subsequent tasks remain unchanged. If the user spends a longer time on a certain task, the control device may mark the difficulty of the content of the task as "moderate", but will not immediately adjust the difficulty or order of subsequent tasks. In addition, the completion status of the task will be simply recorded in the user's historical learning feature record.

[0035] The embodiment can dynamically generate the best learning mode matching the user's cognitive mode and the characteristics of the knowledge content by obtaining the identification information of the user to determine the learning habit of the user and obtaining the historical learning feature record in combination with the knowledge source description. Thus, the generation of the learning task sequence is no longer limited to static presetting, but can be customized according to the personalized needs of the user. In addition, by obtaining the execution feedback of the user on the learning task sequence in real time, the control device can adjust the priority and content difficulty of the learning task, thereby realizing continuous optimization of the learning path, effectively solving the deficiencies of the traditional learning machine in dynamic generation and optimization of the personalized learning path, and improving the adaptability and efficiency of learning.

[0036] The embodiment can dynamically generate the best learning mode matching the user's cognitive mode and the characteristics of the knowledge content by obtaining the identification information of the user to determine the learning habit of the user and obtaining the historical learning feature record in combination with the knowledge source description. Thus, the generation of the learning task sequence is no longer limited to static presetting, but can be customized according to the personalized needs of the user. In addition, by obtaining the execution feedback of the user on the learning task sequence in real time, the control device can adjust the priority and content difficulty of the learning task, thereby realizing continuous optimization of the learning path, effectively solving the deficiencies of the traditional learning machine in dynamic generation and optimization of the personalized learning path, and improving the adaptability and efficiency of learning.

[0037] In practical applications, how to deeply mine these information to ensure that the generated best learning mode can accurately reflect the personalized cognitive mode and learning preference of the user, and avoid generalized or unsuitable learning strategies.

[0038] To this end, with reference to Figure 2 , another embodiment of the present application provides a control method of a learning machine, based on the above-mentioned Figure 1 embodiment, the best learning mode corresponding to the user is generated according to the learning habit and the historical learning feature record, including steps S310-S340, wherein: S310, extract a multi-dimensional behavior feature vector of the user in learning different knowledge types in the historical learning feature record, the multi-dimensional behavior feature vector including a knowledge point switching frequency, a single knowledge point residence time length distribution, and an exercise accuracy rate change pattern over time; S320, obtain user long-term preference information in a learning habit, the user long-term preference information including a selection tendency for graphical, sequenced, or auditory content; S330, generate a comprehensive learning feature portrait according to the multi-dimensional behavior feature vector and the user long-term preference information; S340, match a preset learning mode template library based on the comprehensive learning feature portrait, and determine an optimal learning mode most suitable for a current knowledge source description of the user.

[0039] The multi-dimensional behavior feature vector is a collection of quantification and abstraction of various behavior patterns exhibited by the user in the learning process. For example, the knowledge point switching frequency can reflect the activity or concentration of the user in jumping between different knowledge themes; the single knowledge point residence time length distribution reveals the input degree or understanding difficulty of the user to specific knowledge content; and the exercise accuracy rate change pattern over time directly reflects the learning effect and knowledge mastery curve of the user. These feature vectors can be obtained by real-time collection and analysis of the user's interaction data (such as clicking, browsing, answering, pausing, etc.) on the learning machine, providing objective data support for subsequent user learning state evaluation.

[0040] The user long-term preference information refers to the tendency of the user to a specific learning content presentation form or interaction mode formed in a long time learning practice and relatively stable. For example, the user may prefer to learn through graphs, pictures, mind maps, or tend to learn through linear, step-by-step, logically rigorous text or video, or prefer to acquire knowledge through audio, explanation, podcast, etc. These preference information can be obtained through the user's historical selection behavior on the learning machine, active setting or questionnaire survey, reflecting the user's subjective learning style and habit.

[0041] The weighted fusion of the multi-dimensional behavior feature vector means that different weights are given to different behavior features according to their relative importance in evaluating the user's learning state, and then these weighted features are combined. For example, statistical methods or machine learning algorithms can be used to determine the weights, so that some features that better reflect the user's current learning bottleneck or advantage are highlighted, thereby forming a more representative behavior feature set.

[0042] The generation of a comprehensive learning feature profile involves integrating and abstracting a weighted, multi-dimensional behavioral feature vector with long-term user preference information to form a unified view that comprehensively describes a user's learning ability, preferences, habits, and other dimensions. This profile can be a high-dimensional vector representation or structured data containing multiple key indicators, aiming to provide a unified view of user learning characteristics and serving as the core basis for matching the best learning methods. The preset learning method template library is a collection of various pre-designed learning strategies or patterns tailored to different learning scenarios and user characteristics. Each template in the library can define a complete learning process, including knowledge presentation formats (such as tree diagrams, categorization memorization, and storytelling), interaction methods, feedback mechanisms, and difficulty adjustment strategies. These templates can be designed by education experts or derived through the analysis and abstraction of numerous successful learning cases, providing a wealth of choices for generating personalized learning solutions.

[0043] Determining the optimal learning method involves calculating the similarity between the comprehensive learning feature profile and the features of various templates in a pre-defined learning method template library, or using a classifier for prediction, to find the solution that best matches the user's current learning needs and cognitive patterns within the template library. The matching process also considers the characteristics of the current knowledge source, such as knowledge type, difficulty, and structure, to ensure that the selected learning method is both compatible with the user's characteristics and suitable for the knowledge content.

[0044] Through the above technical solution, this embodiment overcomes the limitation of relying solely on generalized learning habits and historical records to accurately generate personalized learning methods. Specifically, by deeply analyzing the multi-dimensional behavioral feature vectors of users during the learning process of different knowledge types, and combining them with their long-term preference information, a refined and comprehensive learning feature profile is constructed, resulting in a more comprehensive and in-depth understanding of users' cognitive patterns and learning styles. Based on this, precise matching of this profile with a preset learning method template library ensures that the generated optimal learning method not only considers the user's behavioral performance but also takes into account their subjective preferences, thereby significantly improving the adaptability and personalization of the learning method. This enables the learning machine to provide users with truly tailored learning strategies, effectively improving learning efficiency and user satisfaction.

[0045] In practical applications, if the compatibility between the user's deep cognitive patterns and the inherent structural characteristics of the current knowledge source is not fully considered, the generated optimal learning method may not achieve the best learning effect, resulting in limited learning efficiency.

[0046] Therefore, referring to Figure 3 Another embodiment of the present invention provides a control method for a learning machine, based on the above. Figure 2The illustrated embodiment generates a comprehensive learning feature profile based on multi-dimensional behavioral feature vectors and long-term user preference information, including steps S331-S333, wherein: S331. Extract the multidimensional behavioral feature vector to represent user learning stability, attention decay rate and knowledge internalization efficiency. S332. Quantify the selection tendency data in users' long-term preference information to generate confidence scores for graphical, serialized, and auditory content; S333: The learning stability, attention decay rate, and knowledge internalization efficiency are matrix-fused with the corresponding confidence scores to generate a comprehensive learning feature profile.

[0047] Based on a comprehensive learning feature profile, a preset learning method template library is matched to determine the best learning method that best suits the user's current knowledge source description, including steps S341-S343, where: S341. Perform deep semantic analysis on the knowledge source description to generate a corresponding lightweight knowledge graph. The lightweight knowledge graph includes core concept nodes, logical relationships between concepts, and attribute tags of the knowledge domain. S342. Align the comprehensive learning feature profile with the lightweight knowledge graph across dimensions to determine the fit score between the user's cognitive pattern and the knowledge structure features. S343. Based on the fit score, retrieve the optimal matching item from the preset learning method template library to determine the best learning method. The learning method corresponding to the optimal matching item includes dynamically adjusted knowledge presentation rhythm, media combination strategy and interactive feedback mechanism.

[0048] In extracting multidimensional behavioral feature vectors to characterize user learning stability, attention decay rate, and knowledge internalization efficiency, deeper learning ability indicators can be extracted from users' historical learning behavior data. Learning stability can be assessed by analyzing the volatility of knowledge point switching frequency; for example, frequent switching of knowledge points within a short period may indicate poor learning stability. Attention decay rate can be inferred from the distribution of dwell time on a single knowledge point; for example, a long dwell time followed by a rapid decrease may indicate rapid attention decay. Knowledge internalization efficiency can be measured by the change pattern of practice accuracy over time; for example, a consistently and steadily improving accuracy indicates high knowledge internalization efficiency. These indicators are calculated through statistical analysis, trend fitting, or machine learning models of the data stream regarding knowledge point switching frequency, dwell time distribution, and practice accuracy change patterns in the multidimensional behavioral feature vectors, providing refined input for the subsequent generation of a comprehensive learning feature profile.

[0049] In quantifying the selection tendency data in the user's long-term preference information and generating the confidence scores of the graphical, serialized, and auditory content, the user's preference for different media forms (such as graphical, serialized, and auditory) can be converted from qualitative description to quantifiable indicators. The confidence scores can be based on statistical analysis of the user's click rate, completion rate, active selection frequency, and dwell time for each type of media resource in historical learning, or obtained through user questionnaires, implicit feedback, etc. For example, if the user's completion rate for video courses (auditory and serialized) is much higher than that of text materials, the confidence scores of the auditory and serialized content will be higher. These confidence scores reflect the user's learning comfort and acceptance in different presentation forms, providing quantitative basis for subsequent feature fusion.

[0050] In matrix fusion of learning stability, focus decay rate, and knowledge internalization efficiency with corresponding confidence scores to generate a comprehensive learning feature portrait, matrix fusion is a method of structurally integrating different dimensional features. Here, learning stability, focus decay rate, and knowledge internalization efficiency can be used as row vectors, and the confidence scores of graphical, serialized, and auditory content as column vectors. Through matrix multiplication, weighted averaging, or other multi-modal fusion algorithms, these heterogeneous data can be mapped to a unified feature space. For example, a weighting matrix can be designed so that when the user's focus decays quickly, the preference weight for serialized content is higher, to recommend a shorter and more structured learning path. This fusion method can more comprehensively and finely characterize the user's comprehensive learning features, forming a multi-dimensional and high-density comprehensive learning feature portrait.

[0051] In deep semantic analysis of the knowledge source description to generate a corresponding lightweight knowledge graph, the lightweight knowledge graph includes core concept nodes, logical relationships between concepts, and attribute labels of knowledge domains. Deep semantic analysis aims to understand the deep meaning and structure of the knowledge source description. This can be achieved through natural language processing (NLP) techniques, such as named entity recognition (NER) to identify core concept nodes, relation extraction to identify logical relationships between concepts (such as "is", "contains", "causes", etc.), and text classification or topic modeling to determine the attribute labels of knowledge domains (such as "mathematics", "physics", "programming", etc.). "Lightweight" means that the knowledge graph focuses on the core content of the current knowledge source, avoiding excessive expansion, so as to quickly build and match. For example, for the description of "Newton's Second Law", the core concept node "Newton's Second Law" can be parsed, the related concept nodes "force", "mass", and "acceleration" can be parsed, and the logical relationships between them such as "definition" and "influence" can be parsed, and the knowledge domain attribute labels "physics" and "classical mechanics" can be determined.

[0052] In the cross-dimensional alignment of the comprehensive learning feature portrait and the lightweight knowledge graph, the cross-dimensional alignment is the process of comparing and matching the cognitive characteristics of the user portrait with the structural characteristics of the knowledge graph. This can be achieved through various algorithms, such as similarity calculation based on vector space model, embedding both the user portrait and the knowledge graph into the same high-dimensional vector space, and then calculating the cosine similarity between them; or through methods such as graph neural network (GNN), directly matching features on the graph structure. The adaptation score reflects the degree of fit between the user's current cognitive mode (such as learning stability, concentration, preference) and the internal structure of the knowledge source (such as concept complexity, logical depth, media suitability). For example, if the user's concentration decays quickly and prefers graphical content, while the knowledge graph shows that the concept of this knowledge point is complex and the logical chain is long, the adaptation score may be low.

[0053] In the retrieval of the optimal matching item from the preset learning mode template library according to the adaptation score, the determination of the best learning mode, the learning mode corresponding to the optimal matching item contains dynamically adjusted knowledge presentation rhythm, media combination strategy and interactive feedback mechanism. The adaptation score is the key basis for selecting the best learning mode. The preset learning mode template library stores various learning strategies for different cognitive modes and knowledge structures. The control device will retrieve the learning mode template with the highest score or meeting a certain threshold from the template library according to the calculated adaptation score. For example, if the adaptation score indicates that the user needs more detailed guidance and more frequent feedback, the control device may select a template containing "step-by-step explanation, multimedia assistance, and immediate quizzes". The optimal matching item is not only a static template, but also contains dynamically adjusted mechanisms, such as adjusting the rhythm (speed) of knowledge presentation, media combination strategy (text, video, interactive proportion) and interactive feedback mechanism (prompt, encouragement, error correction) according to the user's real-time performance.

[0054] By the technical solution, in generating the comprehensive learning feature portrait, the embodiment not only considers the multi-dimensional behavior features and long-term preferences of the user, but also deeply extracts deep cognitive indexes such as learning stability, concentration decay rate and knowledge internalization efficiency, and performs matrix fusion on the deep cognitive indexes and the quantified media preference certainty score, so as to construct a more fine and comprehensive user comprehensive learning feature portrait. In addition, through deep semantic analysis on the knowledge source description, a lightweight knowledge graph is generated, so that the control device can understand the internal structure and complexity of the knowledge. On this basis, the fine user comprehensive learning feature portrait and the knowledge graph are cross-dimensionally aligned, and an adaptation score between the user cognitive mode and the knowledge structure features is calculated, so that the degree of fit between the user and the current knowledge content can be more accurately evaluated. Finally, based on the adaptation score, the optimal matching item is retrieved from the preset learning mode template library, which not only provides the best learning mode, but also includes a dynamically adjusted knowledge presentation rhythm, a media combination strategy and an interactive feedback mechanism. This makes the generated best learning mode more accurately adapt to the individual cognitive characteristics of the user and the structural characteristics of the current knowledge, effectively solves the problem of insufficient matching degree between the learning mode and the knowledge content in the traditional method, significantly improves the individualization degree and efficiency of learning, and ensures that the user can efficiently master the knowledge in the most suitable way for himself.

[0055] When converting the abstract knowledge source description into specific and executable learning task sequences, how to ensure the effective organization of knowledge content, adapt to the cognitive characteristics of the user and provide appropriate feedback mechanism To this end, with reference to Figure 4 , a control method of a learning machine is provided, based on the above Figure 1 The embodiment shown generates a corresponding learning task sequence according to the best learning mode, including steps S410-S450, wherein: S410, calling a knowledge structuring conversion rule corresponding to the best learning mode; S420, converting the knowledge content corresponding to the knowledge source description according to the knowledge structuring conversion rule to generate a knowledge node graph with a logical hierarchy; S430, according to the single learning time and rhythm preference in the learning habit corresponding to the user, the knowledge node graph is divided into a plurality of learning task blocks, each learning task block has independent logical coherence; S440, generating an executable learning task and a feedback node corresponding to each learning task block, the feedback node includes an embedded selection question, a knowledge repetition trigger point or a scenario simulation operation; S450, sorting the executable learning tasks according to the priority to generate a learning task sequence.

[0056] Among them, calling the knowledge structure conversion rule corresponding to the best learning mode refers to that the system selects and activates the corresponding knowledge processing logic from the preset rule library according to the currently determined best learning mode (for example, tree diagram, classification memory method, listening to stories, etc.). These rules are a set of algorithms or templates designed to guide how to structure the original knowledge content. For example, if the best learning mode is "tree diagram", the called rule may focus on identifying core concepts, extracting sub-concepts, establishing hierarchical relationships, etc.; if it is "classification memory method", the rule may involve keyword extraction, classification label generation, correlation analysis, etc.

[0057] Subsequently, the corresponding knowledge content of the knowledge source description is converted according to the knowledge structure conversion rule to generate a knowledge node graph with a logical hierarchy. This process aims to transform the original, possibly unstructured or semi-structured knowledge content (such as text, video scripts, data, etc.) into a structured, easily machine-processed and user-understood graph form. The conversion process can use natural language processing (NLP) techniques to perform semantic analysis on text, identify entities, relationships, and events; for multimedia content, it can combine speech recognition and image recognition techniques to extract key information. The generated knowledge node graph can be represented as a collection of nodes (representing concepts, facts, skill points) and edges (representing relationships between concepts such as "contains", "cause and effect", "parallel", etc.), forming a directed acyclic graph or more complex network structure, ensuring the logical coherence and hierarchy of knowledge.

[0058] On this basis, according to the user's single learning duration and rhythm preference in the corresponding learning habit, the knowledge node graph is divided into multiple learning task blocks, each learning task block has independent logical coherence. This step is to divide the huge knowledge graph into units suitable for user short-term learning and digestion. The single learning duration preference (for example, the user tends to learn for 20-30 minutes each time) and the rhythm preference (for example, the user prefers to learn theory first and then practice, or interspersed with exercises) in the learning habit are key inputs. The control device can intelligently divide the boundaries according to these preferences, combined with the logical structure of the knowledge node graph. For example, a learning task block can be a complete section, a concept cluster, or a skill module, ensuring the integrity and logical consistency of the internal knowledge points, and avoiding knowledge fragmentation. The segmentation algorithm can consider the density of nodes in the graph, the strength of the relationship between concepts, and the user-set time limit.

[0059] Wherein, by converting abstract knowledge blocks into specific, user-operable and interactive learning activities, and integrating instant feedback mechanisms. For each learning task block, the system will automatically generate corresponding learning tasks according to its content type and learning objectives. For example, if the block is concept understanding, reading tasks, video watching tasks can be generated; if the block is skill mastery, programming exercises, simulation operation tasks can be generated. Feedback nodes are key components in learning tasks, used to assess the user's understanding and mastery. Embedded multiple-choice questions can appear immediately after knowledge point explanation; knowledge repetition triggers can prompt users to summarize the learned content in their own words; scenario simulation operations provide a virtual environment for users to apply what they have learned. The design of these feedback nodes should be closely integrated with learning methods and knowledge types, providing diverse forms of interaction.

[0060] Finally, by providing users with a clear, orderly learning path, ensuring the progressive and effectiveness of learning. Priority ranking can be based on multiple factors: the prior and posterior dependency relationship of knowledge points (for example, basic concepts must be mastered before learning advanced concepts), the user's weak points in historical learning (prioritize reinforcement exercises), the rhythm preferences in learning habits (for example, easy first or alternating theory and practice). The sorting algorithm can combine topological sorting, weighted scoring and other methods to ensure that the generated learning task sequence not only meets the knowledge logic, but also maximizes the adaptation to the user's individualized needs and learning efficiency.

[0061] The embodiment can convert the knowledge content corresponding to the knowledge source description into a knowledge node graph with logical hierarchy by calling the knowledge structure conversion rules corresponding to the best learning method, thereby ensuring the systematicness and completeness of the knowledge content. On this basis, according to the user's single learning duration and rhythm preferences in learning habits, the knowledge node graph is intelligently segmented into multiple learning task blocks with independent logical coherence, effectively avoiding knowledge fragmentation and improving the user's absorption efficiency of learning content. By generating corresponding executable learning tasks and diverse feedback nodes for each learning task block, such as embedded multiple-choice questions, knowledge repetition triggers or scenario simulation operations, the interactivity and immediacy of learning are greatly enhanced, allowing users to promptly check learning effectiveness and obtain targeted guidance. Finally, the executable learning tasks are prioritized to form a personalized learning task sequence, not only optimizing the learning path, but also significantly improving the relevance and effectiveness of learning, thereby solving the challenge of converting abstract knowledge into an efficient and personalized learning sequence.

[0062] In practical applications, how to efficiently and accurately obtain the historical learning feature records highly related to the current input knowledge source description of the user, especially when the local data is insufficient or the knowledge source description is relatively novel, is a problem to be solved. If the historical learning feature records are not obtained in time or comprehensively, the accuracy and efficiency of the subsequent optimal learning mode generation may be affected.

[0063] Optionally, with reference to Figure 5 The present application also provides a control method of a learning machine, based on the above Figure 1 According to the embodiment shown in the figure, in response to the typing operation for the knowledge source description, the historical learning feature records corresponding to the knowledge source description are obtained, including steps S210-S240, wherein: S210, in response to the knowledge source description input by the user terminal, the key theme words and the main body classification of the knowledge source description are analyzed, the subject field and the knowledge dimension to which the knowledge source belongs are determined; S220, based on the subject field and the knowledge dimension, the historical learning feature records associated with the knowledge source in the local cache are searched; S230, if no matching record is searched, a communication connection with the cloud server is established; S240, the historical learning feature records matching the knowledge source description in the cloud server are obtained.

[0064] The present embodiment first performs semantic understanding and structural processing on the unstructured or semi-structured knowledge source description input by the user. Among them, the learning machine can use natural language processing (NLP) technology, such as keyword extraction, text classification, named entity recognition, etc., to identify the core concept, theme and its belonging to a wider subject field (such as mathematics, physics, history, etc.) and knowledge dimension (such as basic theory, application practice, concept analysis, etc.) from the text input by the user. For example, when the user inputs “derivation process of Newton's second law”, the system can analyze that “Newton's second law” is the key theme word, “physics” is the subject field, and “derivation process” is the knowledge dimension. In this way, the system can convert the user's vague intention into precise query conditions for machine processing.

[0065] Secondly, after determining the subject field and knowledge dimension of the knowledge source, the control device will prioritize searching in the locally stored cache data. The local cache usually contains the user's recent or frequently used knowledge points learning history data, which has fast access speed and can provide instant response. The retrieval process can use the pre-established indexing mechanism, such as inverted index or hash table, according to the parsed subject field and knowledge dimension as the query key, to quickly find out whether there is a historical learning feature record related to the current knowledge source description (or a knowledge source with similar semantics) in the local. For example, if the user has previously learned "Newton's Law of Motion", and there is a related record in the local cache, the control device will preferentially call these data.

[0066] Furthermore, when the local cache fails to find a historical learning feature record that matches the current knowledge source description, or the richness of the local record is insufficient to support the subsequent personalized learning mode generation, the learning machine will automatically determine and start the communication connection with the cloud server. This usually involves establishing a secure and stable data transmission channel with the pre-set remote server through the network interface (such as Wi-Fi, cellular data, etc.) to expand the data source.

[0067] Finally, once the connection with the cloud server is established, the learning machine will send a query request containing the parsed knowledge source description (key topic words, subject field, knowledge dimension, etc.) to the cloud server. The cloud server usually has a more extensive and comprehensive user learning behavior database, including the user's learning history on different devices, as well as a large amount of anonymous learning data of other similar users. The cloud server will use its powerful computing and storage capabilities to perform complex distributed retrieval and big data analysis algorithms to filter out the most matching historical learning feature records from the massive data, and transmit them back to the learning machine. These records may include the user's learning duration, exercise performance, preferred media form, etc. for this knowledge point or related knowledge points.

[0068] By the technical solution, the embodiment can firstly perform deep analysis on the knowledge source description input by the user, accurately identify the key theme and the field to which the user belongs, and thus provide accurate positioning basis for subsequent data retrieval. On this basis, the control device preferentially performs fast retrieval in the local cache, fully utilizes the existing and responsive data resources, and effectively improves the efficiency of obtaining the historical learning feature records. When the local data is insufficient to support subsequent decision-making, the control device can intelligently extend to the cloud server to obtain more comprehensive and rich historical learning feature records, and make up for the limitations of the local data. This layered and intelligent retrieval mechanism ensures that whether the knowledge source description is common or novel, the historical learning feature records highly related to the user and the knowledge source can be efficiently and accurately obtained, thereby providing a solid data foundation for subsequent generation of the optimal learning mode of the user and optimization of the learning path, and significantly improving the personalization and effectiveness of the learning experience.

[0069] Although the priority and content difficulty of the learning task sequence can be adjusted according to the execution feedback of the user on the learning task sequence to optimize the learning path of the user, if only rough adjustment is performed, deep problems in the learning process of the user, such as weak links in knowledge mastery, fluctuations in understanding efficiency, and shifts in interest points, can not be accurately identified, thereby resulting in poor adjustment effect and difficulty in truly realizing personalized and efficient optimization of the learning path.

[0070] Therefore, with reference to Figure 6 , another embodiment of the present application provides a control method of a learning machine, which is based on the above-mentioned Figure 1 According to the execution feedback, the priority and content difficulty of the learning task sequence are adjusted to optimize the learning path of the user, and the historical learning feature records are updated, including steps S610-S650, wherein: S610, analyzing the execution feedback to determine the knowledge weak points, understanding speed abnormal points, and interest attenuation points shown by the user in the learning task sequence; S620, based on the knowledge weak points and the understanding speed abnormal points, determining the content difficulty adjustment amount of the corresponding task node in the learning task sequence, and the content difficulty adjustment amount includes increasing the pre-laid information, replacing the explanation perspective, or decomposing the complex steps; S630, based on the interest attenuation points, reevaluating the priority of the tasks not yet performed in the learning task sequence, and generating a corresponding task priority adjustment strategy; S640, according to the content difficulty adjustment amount and the task priority adjustment strategy, reconstructing the learning task sequence to optimize the learning path of the user; S650, updating the optimized learning path to the historical learning feature records.

[0071] The embodiment first analyzes the execution feedback to determine the weak knowledge points, abnormal understanding speed points, and interest attenuation points exhibited by the user in the learning task sequence. The control device deeply analyzes all interaction data generated by the user during the execution of the learning task sequence, which may include the user's completion time on a specific task, the accuracy of the answers, the number of repeated viewing or reading, the skipped or quickly browsed content, and the duration of staying on specific content, etc. Through data mining and machine learning algorithms, the control device can identify the weak knowledge points where the user lacks understanding of specific knowledge points or concepts, such as frequent errors in related test questions; identify the abnormal understanding speed points where the user's understanding speed deviates significantly from the average level when processing specific learning content, such as staying too long or too short on certain tasks; and identify the interest attenuation points where the user's interest in specific learning content or learning tasks decreases, which may be reflected through the user's reduced interaction frequency, distracted attention, and other behavioral patterns on related tasks.

[0072] Based on the determined weak knowledge points and abnormal understanding speed points, the control device determines the content difficulty adjustment amount of the corresponding task nodes in the learning task sequence. The content difficulty adjustment amount may include adding pre-laid information, changing the explanation perspective, or decomposing complex steps. For example, when it is identified that the user has weak points in a certain knowledge point, more basic and understandable preparatory knowledge or background information can be supplemented before or inside the knowledge point task to help the user build a bridge for understanding; when the user has difficulty understanding a certain content, the knowledge point can be re-explained from different angles, using different metaphors or cases, such as switching from theoretical explanation to practical application cases; for abnormal understanding speed points, especially when the user has difficulty in processing complex tasks, a large and complex learning task or knowledge point can be decomposed into multiple smaller and more digestible sub-tasks or steps to reduce the cognitive load of single learning.

[0073] In addition, based on the determined interest attenuation points, the control device re-evaluates the priority of the tasks in the learning task sequence that have not been executed and generates a corresponding task priority adjustment strategy. The strategy aims to improve the priority of tasks that are strongly associated with the user's interest points and reduce the priority of secondary derivative tasks. For example, if the user shows high interest in a certain topic or form of content, the control device will prioritize subsequent tasks that are closely related to the interest points to maintain and stimulate the user's learning motivation; while those tasks that have low relevance to the user's current interest points or belong to auxiliary and expanding tasks, their priority will be appropriately reduced to avoid weakening the user's interest due to forced learning.

[0074] Subsequently, according to the content difficulty adjustment amount and the task priority adjustment strategy, the control device will reconstruct the learning task sequence to optimize the user's learning path. The reconstruction process involves generating a new learning task sequence that better meets the user's current learning state and needs based on the above-mentioned adjustment amount and strategy on the basis of the original learning task sequence, which may include operations such as task insertion, deletion, sequence adjustment, content modification, etc. Finally, the optimized learning path is updated to the historical learning feature record, ensuring that the control device can continuously learn and adapt to the user's changes, providing more accurate basis for future learning path generation and adjustment.

[0075] Through the above technical solutions, the embodiment can finely analyze the user's execution feedback, thereby accurately identifying the user's weak points in knowledge, abnormal points in understanding speed, and points of interest decay in the learning process. Based on these detailed insights, the control device can adjust the content difficulty of the learning task, such as adding pre-laid information, changing the explanation perspective, or breaking down complex steps, to effectively solve the user's understanding obstacles at specific knowledge points. By dynamically adjusting the priority of tasks that have not been executed, prioritizing tasks that are strongly associated with the user's interest, and reducing the priority of secondary tasks, the user's learning interest and motivation can be effectively stimulated and maintained. This dual adjustment mechanism combining content difficulty and task priority makes the reconstruction of the learning task sequence more accurate and personalized, thereby significantly optimizing the user's learning path and improving learning efficiency and user satisfaction. In addition, updating the optimized learning path to the historical learning feature record ensures that the system can continuously iterate and adapt to the user's long-term learning needs, achieving truly adaptive learning.

[0076] In practical applications, how to present complex learning task sequences in an intuitive and effective manner, and accurately and meticulously capture the user's real behavior and state in the learning process, to ensure that subsequent learning path optimization can accurately reflect the user's learning needs and performance, is a problem that needs to be solved.

[0077] To this end, with reference to Figure 7 , another embodiment of the present application provides a control method of a learning machine, based on the above-mentioned Figure 1 The learning machine includes a display component; the learning task sequence is displayed, and the user's execution feedback on the learning task sequence is obtained, including steps S510-S540, wherein: S510, control the display component to render and display the learning task sequence in the form of a process diagram or a list; S520, in response to a selection operation on any task node in the learning task sequence, call the learning content corresponding to the task node; S530, in response to the interactive operation for the learning content, collecting time sequence data of the user in content browsing, pausing, backtracking, accelerating, and answering behaviors, etc. S540, generating an execution feedback based on the time sequence data, the execution feedback including task completion degree, interaction frequency, and abnormal interval of stay time.

[0078] Among them, the display component in the learning machine is configured to be able to present the learning task sequence in multiple visual forms, aiming to improve the user's understanding and sense of control of the learning path. When rendered in the form of a process diagram, the learning task sequence can be displayed as a flowchart, a mind map, a timeline, or a progress bar, etc. These graphical representations can intuitively reveal the logical relationship, time sequence, or overall completion degree between tasks. For example, the flowchart can clearly show the dependency of tasks, and the mind map highlights the hierarchical structure of knowledge points. When rendered in the form of a list, the learning task sequence is presented in an itemized manner, each item can contain task name, estimated time consumption, difficulty level, completion status, etc. Detailed information, users can expand or fold according to needs, in order to facilitate review and management.

[0079] When the user selects a certain task node in the learning task sequence by clicking, touching, voice command or other interaction methods, the control device will identify the operation, and retrieve and load the teaching resources associated with the task node from the storage medium (such as local storage or cloud server), such as text, pictures, videos, audio, interactive exercises or simulation environment, etc., so as to call the learning content corresponding to the task node.

[0080] The control device will continuously monitor all interactive behaviors of the user with the learning content, and collect time sequence data of the user in content browsing, pausing, backtracking, accelerating, and answering behaviors, etc. For example, in terms of content browsing, the control device will record the start time, end time, scrolling speed, page stay time, etc. of the user watching videos, reading texts; in terms of pausing, backtracking, accelerating, the control device will record the time points and duration of the user's pause, fast forward, fast backward operations in video or audio content, and the multiple of accelerated playback; in terms of answering, the control device will record the answering time, answer selection, submission time, modification times, etc. of the user in multiple-choice questions, fill-in-the-blank questions, question-and-answer questions, etc. Exercises. These data are stored in the form of time sequence, so as to analyze the dynamic changes of user behavior subsequently.

[0081] Based on the collected time-series data, the control device generates execution feedback. This feedback includes task completion rate, interaction frequency, and abnormal dwell time intervals. Task completion rate can be determined by calculating the proportion of completed task nodes to the total number of task nodes, or by the percentage of sub-tasks or learning content completed by the user within each task node, such as the completion rate of watching videos, reading articles, or the accuracy rate of practice questions. Interaction frequency refers to the total number of clicks, swipes, inputs, pauses, and rewinds performed by the user on specific learning content or task nodes, reflecting the user's activity and engagement. Abnormal dwell time intervals are identified by analyzing the time users spend on specific knowledge points or learning content and comparing it with the average learning time for that knowledge point or the user's historical learning patterns. Significantly long or short dwell time intervals are identified; for example, if a user's dwell time on a particular knowledge point far exceeds the average, it may indicate that that knowledge point is a weak point or a difficult area for the user.

[0082] Through the above technical solution, this embodiment can present the learning task sequence to the user in an intuitive process diagram or list format, enabling the user to clearly understand the learning progress and task structure. By responding to the user's selection of task nodes to retrieve learning content, and meticulously collecting time-series data on various user interactions such as content browsing, pausing, rewinding, accelerating, and answering questions, this embodiment can generate multi-dimensional execution feedback including task completion rate, interaction frequency, and abnormal intervals in dwell time. This refined feedback mechanism allows the control device to more accurately understand the user's learning status, comprehension level, and potential difficulties, providing solid data support for subsequently adjusting the priority and content difficulty of the learning task sequence based on execution feedback. This allows for more precise optimization of the user's learning path, significantly improving the personalization and effectiveness of learning.

[0083] If the methods for acquiring learning habits are too simplistic or based on limited data, they may not accurately reflect the user's true learning preferences and cognitive patterns, thus affecting the generation of the optimal learning methods and the personalization of the learning task sequence.

[0084] Therefore, referring to Figure 8 Another embodiment of the present invention provides a control method for a learning machine, based on the above. Figure 1 The illustrated embodiment obtains the user's identification information and determines the user's corresponding learning habits, including steps S110-S130, wherein: S110. Obtain the user's unique identification information through the user login credentials and access the cloud user profile database; S120, retrieve the long-period historical learning summary report of the user from the cloud user archive, the long-period historical learning summary report including the click rate, completion rate and active search preference of the user on different media form learning resources; S130, based on the long-period historical learning summary report, construct a multi-dimensional learning preference portrait of the user, identify the tendency of the user to text, video or interactive content, and generate learning habits.

[0085] Wherein, the unique identification information of the user is obtained through the user login credential, and the cloud user archive is accessed to establish the uniqueness of the user's identity, and the long-term accumulated learning data of the user is obtained based on this. The user login credential can be a username and password, a mobile phone verification code, a third-party authorized login and the like. Its function is to verify the legality of the user's identity. Once the identity verification is passed, the control device can obtain the identification information uniquely bound to the user, such as user ID. Subsequently, the control device uses the unique identification information to safely access the user archive stored in the cloud. The cloud user archive is a database that centrally stores user personal information, historical learning data, preference settings and the like. Its advantages lie in the persistence, scalability and convenience of cross-device access of data storage, ensuring that the user can obtain consistent and complete personal learning data when using the learning machine anywhere.

[0086] On this basis, the long-period historical learning summary report of the user is retrieved from the cloud user archive, the long-period historical learning summary report including the click rate, completion rate and active search preference of the user on different media form learning resources. The long-period historical learning summary report is not a simple short-term record, but a collection of user learning behavior data over a long period of time (for example, several months or years). The report records in detail the user's interaction with different media form learning resources (such as pure text, text and pictures, video explanation, audio course, interactive exercise, etc.). Among them, "click rate" reflects the user's interest and selection tendency for a particular type of content; "completion rate" reveals the user's perseverance and learning perseverance when facing different difficulty or form content; "active search preference" reflects the user's interest points and learning strategies when exploring knowledge independently. Through these multi-dimensional data, the system can comprehensively and objectively evaluate the user's learning habits, rather than relying solely on the user's subjective filling or short-term performance.

[0087] Among them, based on the long-period historical learning summary report, a user multi-dimensional learning preference portrait is constructed, the tendency of the user to text, video or interactive content is identified, and learning habits are generated to convert the original historical learning data into structured and analyzable user preference information. Through comprehensive analysis and modeling of the click rate, completion rate, and active search preference data in the long-period historical learning summary report, the control device can construct a multi-dimensional user learning preference portrait. For example, if the user's click rate and completion rate of video courses are much higher than that of pure text content, and the user tends to search for resources with keywords such as "explanation" and "demonstration" when actively searching, it can be identified that the user has a strong tendency for video content. Similarly, the user's tendency for text, interactive content, etc. can be identified. This portrait not only quantifies the user's preference for different media forms, but also may contain implicit preferences for learning pace, difficulty, feedback method, etc. Ultimately, these identified tendencies are integrated and abstracted to form the user's corresponding "learning habits", providing accurate input for subsequent generation of the best learning method.

[0088] Through the above technical solutions, when obtaining the identification information of the user and determining the learning habits of the user, the embodiment no longer relies on simple user input or short-term behavior, but ensures the uniqueness of the identity through the user login credential and securely accesses the cloud user archive. On this basis, the control device can retrieve and deeply analyze the rich data such as click rate, completion rate, and active search preference contained in the long-period historical learning summary report of the user, thereby comprehensively and objectively reflecting the real interaction mode of the user on different media form learning resources. Based on these multi-dimensional and long-period behavior data, the control device can construct a more accurate user multi-dimensional learning preference portrait, accurately identify the user's tendency for different forms such as text, video or interactive content. This refined learning habit generation method significantly improves the depth and accuracy of understanding of the user's personalized learning needs, and provides a more solid and reliable foundation for subsequent generation of the user's corresponding best learning method based on the learning habits and historical learning feature records, so that the generated best learning method and learning task sequence can better fit the user's actual cognitive mode and preferences, effectively improving learning efficiency and user satisfaction.

[0089] The application also provides a control device, which comprises a memory, a processor, and a control program of a learning machine stored in the memory and executable on the processor, and the control program of the learning machine is configured to implement the control method of the learning machine.

[0090] It is worth noting that, since the control device of the application is based on the control method of the learning machine as described above, the embodiments of the control device of the application include all the technical solutions of all the embodiments of the control method of the learning machine, and achieve the same technical effects, which will not be repeated here.

[0091] With the development of educational informatization and artificial intelligence technology, intelligent learning machines face significant challenges in assisting personalized learning processes. Existing technologies are difficult to deeply integrate users' long-term learning habits, historical learning characteristics in specific knowledge fields, and real-time learning feedback, resulting in the inability to dynamically decide and adapt to the best learning method, while lacking the ability to continuously optimize the priority of learning task sequences, content difficulty, and presentation paths, leading to insufficient learning path adaptability and limited learning efficiency and experience improvement.

[0092] The present application also provides a learning machine, which comprises a display component and a control device according to any one of the above embodiments, wherein: The display component is used to present a learning task sequence and an interactive interface; the control device is electrically connected to the display component; the control device is used to obtain the identification information of a user, determine the learning habits of the user, and in response to a typing operation on a knowledge source description, obtain the historical learning characteristic records corresponding to the knowledge source description, then generate the best learning method corresponding to the user according to the learning habits and the historical learning characteristic records, generate the corresponding learning task sequence according to the best learning method, display the learning task sequence, obtain the execution feedback of the user on the learning task sequence, adjust the priority and content difficulty of the learning task sequence according to the execution feedback, optimize the learning path of the user, and update the historical learning characteristic records.

[0093] The core innovation of the present embodiment lies in the multi-dimensional integration of user learning habits and historical learning characteristic records, combined with a dynamic closed-loop adjustment mechanism based on execution feedback, thereby realizing real-time optimization of the priority and content difficulty of the learning task sequence, and achieving the effect of adaptive evolution of personalized learning paths. Specifically, the control device first determines the long-term learning habits of the user based on the user's identification information, including the tendency to text, video or interactive content and the preference for single learning duration; at the same time, in response to the typing operation on the knowledge source description, the historical learning characteristic records corresponding to the knowledge source are accurately obtained, including key data such as knowledge point switching frequency, residence time distribution and exercise accuracy change pattern. On this basis, the control device generates the best learning method according to the learning habits and the historical learning characteristic records, such as using tree diagrams for deduction for knowledge points with strong logic, or using story listening for situational immersion for abstract concepts, and then structurally converts the knowledge content into a knowledge node graph with logical levels according to the best learning method, and divides it into multiple learning task blocks with independent logical coherence.

[0094] The display component renders the learning task sequence in a process diagram or a list form, the control device collects time sequence data of the user's behaviors in content browsing, pausing, backtracking, and answering, etc. in real time, and generates execution feedback including task completion degree, interaction frequency, and abnormal interval of stay time. Based on the execution feedback, the control device analyzes the user's weak knowledge points, abnormal understanding speed points, and interest decay points, dynamically adjusts the content difficulty (such as adding pre-laid information or decomposing complex steps) and priority (increasing the priority of tasks strongly associated with interest points) of subsequent tasks, thereby reconstructing the learning task sequence and updating the historical learning feature record. Through the above technical solution, the system breaks through the limitations of the traditional static content organization mode, realizes the transformation of the learning path from the preset rules to the dynamic closed-loop optimization, and effectively improves the learning adaptability and efficiency.

[0095] In one implementation of the embodiment, it is assumed that user A is a high school student who wants to understand the physical concept of "quantum entanglement" in depth through the learning machine. First, the learning machine obtains the identification information of user A and determines his learning habits. When user A logs in to the learning machine, the control device obtains the unique identification information through his login credentials and accesses the cloud user archive. From the archive, the control device retrieves the long-term historical learning summary report of user A. The report shows that user A has a high click rate and completion rate for video explanations when learning abstract physical concepts, and actively searches for interactive simulation experiments. Based on this report, the learning machine builds a multi-dimensional learning preference portrait of user A, identifies his strong preference for visual and interactive content, and generates user A's learning habits, such as preferring to learn through animations and simulation experiments, preferring a single learning duration of about 25 minutes, and preferring a medium to fast learning pace.

[0096] Then, user A enters "quantum entanglement" as a knowledge source description on the learning machine. The learning machine responds to this input operation, analyzes the key keywords and main categories of "quantum entanglement", and determines that it belongs to the field of "quantum physics" and the knowledge dimension of "abstract concept understanding". The control device first retrieves the historical learning feature records associated with the knowledge source in the local cache. If no matching record is found locally, the control device will establish a communication connection with the cloud server to obtain the historical learning feature records of user A related to the field of "quantum physics", especially the concept of "quantum entanglement" in the cloud server. The record may show that user A has a slow initial understanding speed when learning similar abstract concepts, but his understanding efficiency is significantly improved through graphical assistance, and he has strong ability to sort out the logical relationship between concepts.

[0097] Subsequently, the learning machine generates the optimal learning mode corresponding to user A according to the learning habits and historical learning feature records of user A. The system extracts the multi-dimensional behavior feature vector of user A when learning different knowledge types in the historical learning feature records, which includes the knowledge point switching frequency (for example, the switching frequency is low in abstract concept learning), the single knowledge point residence time distribution (for example, the residence time of key concepts is longer), and the change pattern of practice accuracy over time (for example, the initial accuracy is low, but it can be steadily improved after a period of learning). At the same time, the system obtains the long-term preference information of the user in the learning habits, that is, the selection tendency of user A to graphical, sequential or auditory content (for example, there is a strong selection tendency to graphical and interactive content). The system analyzes the data flow in the multi-dimensional behavior feature vector about the learning stability, concentration decay rate and knowledge internalization efficiency of the user, and quantifies the selection tendency data in the long-term preference information of the user to generate the confidence score of graphical, sequential and auditory content. The learning stability, concentration decay rate and knowledge internalization efficiency are matrix fused with the corresponding confidence scores to generate the comprehensive learning feature portrait of user A. The portrait shows that user A is suitable for understanding complex concepts through visual aids and logical reasoning. The control device performs deep semantic analysis on "quantum entanglement" to generate a corresponding lightweight knowledge graph, including core concept nodes (such as "entangled state", "Bell inequality"), logical relationships between concepts and attribute labels of the knowledge field. The comprehensive learning feature portrait and the lightweight knowledge graph are cross-dimensionally aligned to determine the adaptation score between the cognitive mode and the knowledge structure characteristics of user A. According to the adaptation score, the control device retrieves the optimal matching item from the preset learning mode template library and determines that "interactive concept map construction" is the optimal learning mode. This way includes dynamically adjusted knowledge presentation rhythm, media combination strategy (such as animation, simulation experiment) and interactive feedback mechanism. Unlike existing technologies that only provide preset text or video, the present method can dynamically match and generate highly customized learning modes according to the personalized characteristics of user A.

[0098] Next, the learning machine generates a corresponding learning task sequence according to the best learning mode of "interactive concept map construction". The control device calls the knowledge structuring conversion rule corresponding to this learning mode, converts the knowledge content corresponding to "quantum entanglement", and generates a knowledge node map with a logical hierarchy. For example, "quantum entanglement" is decomposed into "basic concept", "experimental verification", "application prospect" and other core nodes, and the hierarchy and association between them are established. According to the single learning duration (25 minutes) and rhythm preference in user A's learning habits, the control device divides the knowledge node map into multiple learning task blocks, for example, the first block is "quantum entanglement basic concept", the second block is "Bell inequality and experimental verification", and each block has independent logical coherence. The control device generates executable learning tasks and feedback nodes corresponding to each learning task block. For example, in the "quantum entanglement basic concept" block, there is an animation demonstration task, a concept fill-in-the-blank question (embedded multiple-choice question), and a trigger point that requires the user to repeat the core concept (knowledge repetition trigger point). Finally, the executable learning tasks are prioritized to generate a learning task sequence. Unlike existing systems that usually provide fixed chapters or simple classification, this method can dynamically segment tasks according to user preferences and knowledge structure, and embed diverse feedback nodes in the tasks.

[0099] Then, the learning machine displays the learning task sequence and obtains the user's execution feedback on the learning task sequence. The learning machine controls the display component to render the learning task sequence in the form of a process chart, and user A can see the overall progress and current task of the "quantum entanglement" learning. User A selects the first task node "quantum entanglement basic concept" in the task sequence. The learning machine retrieves the learning content corresponding to this task node, such as an interactive animation. User A watches the animation, fills in the concept blanks, tries to repeat, and other interactive operations, and the learning machine collects the timing data of the user's behavior in content browsing, pausing, backtracking, accelerating, and answering questions. For example, the user pauses and backtracks several times at a complex concept, hesitates for a long time on the fill-in-the-blank question, and makes a mistake the first time. Based on the timing data, the execution feedback is generated. The execution feedback includes task completion degree (e.g. 80%), interaction frequency (e.g. backtracking 3 times, pausing 2 times), and abnormal interval of stay time (e.g. staying time at "superposition state" concept far exceeds the average value).

[0100] Finally, the learning machine adjusts the priority and content difficulty of the learning task sequence according to the execution feedback, optimizes the learning path of the user, and updates the historical learning feature record. The learning machine analyzes the execution feedback, determines the weak knowledge points (inadequate understanding of the "superposition state" concept) and understanding speed abnormal points (long time staying at the "superposition state") and interest attenuation points (long time no operation at a certain theoretical derivation link) shown by the user A in the learning task sequence. Based on the weak knowledge points and understanding speed abnormal points, the learning machine determines the content difficulty adjustment amount of the corresponding task node in the learning task sequence. For example, for the "superposition state" concept, add pre-laid information (such as introducing the concept of probability through life examples), change the explanation perspective (such as switching from mathematical formulas to physical images), or decompose complex steps. Based on the interest attenuation point, the learning machine reevaluates the priority of the tasks in the learning task sequence that have not been executed. For example, increase the priority of tasks that are strongly associated with the interest points (such as interactive simulation experiments) shown by user A, and reduce the priority of secondary derivative tasks (such as pure theoretical derivation). Generate the corresponding task priority adjustment strategy. According to the content difficulty adjustment amount and the task priority adjustment strategy, the learning machine reconstructs the learning task sequence. For example, in the next task block, a simplified interactive simulation of "superposition state" is inserted first, and then the explanation of "Bell's inequality" is entered, and the explanation method of "Bell's inequality" is adjusted to a more visually guided flowchart. Update the optimized learning path to the historical learning feature record of user A. This record will reflect the understanding mode and preference adjustment of user A on the concept of "superposition state" in the field of "quantum physics". Unlike existing systems that only provide repeated exercises or simple prompts when users encounter difficulties, this method can dynamically adjust the difficulty, priority and presentation of subsequent tasks according to real-time feedback, and feed back the results of these dynamic adjustments to the historical learning feature record of the user, achieving closed-loop optimization and continuous adaptation of the learning path.

[0101] It is worth noting that since the learning machine of the present application is based on the above-mentioned control device, the embodiments of the learning machine of the present application include all the technical solutions of all the embodiments of the above-mentioned control device, and the technical effects achieved are also completely the same, which will not be repeated here.

[0102] The above is only the preferred embodiment of the present application, and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation based on the content of the specification and drawings, or direct or indirect application in other related technical fields, is also included in the patent protection scope of the present application.

Claims

1. A control method for a learning machine, characterized in that, The control method of the learning machine includes: Obtain user identification information to determine the user's corresponding learning habits; In response to an input operation for a knowledge source description, retrieve historical learning feature records corresponding to the knowledge source description; Based on the learning habits and historical learning characteristics recorded, the optimal learning method for each user is generated. Generate the corresponding learning task sequence according to the optimal learning method described above; Display the learning task sequence and obtain user feedback on the execution of the learning task sequence; Based on the execution feedback, the priority and difficulty of the learning task sequence are adjusted to optimize the user's learning path and the historical learning feature records are updated.

2. The control method for the learning machine as described in claim 1, characterized in that, The step of generating the optimal learning method for a user based on the recorded learning habits and historical learning characteristics includes: Extract multi-dimensional behavioral feature vectors from the historical learning feature records of users when learning different types of knowledge. The multi-dimensional behavioral feature vectors include the frequency of knowledge point switching, the distribution of dwell time for a single knowledge point, and the change pattern of practice accuracy over time. Obtain long-term user preference information from the learning habits, including the user's preference for graphical, sequential, or auditory content; A comprehensive learning feature profile is generated based on the multidimensional behavioral feature vector and the user's long-term preference information. Based on the comprehensive learning feature profile, a preset learning method template library is matched to determine the best learning method that best matches the user's current knowledge source description.

3. The control method for the learning machine as described in claim 2, characterized in that, The step of generating a comprehensive learning feature profile based on the multidimensional behavioral feature vector and the user's long-term preference information includes: Extract the multidimensional behavioral feature vectors that represent user learning stability, attention decay rate, and knowledge internalization efficiency. Quantify the selection tendency data in the user's long-term preference information to generate confidence scores for graphical, serialized, and auditory content; The learning stability, the attention decay rate, and the knowledge internalization efficiency are matrix-fused with the corresponding confidence scores to generate a comprehensive learning feature profile. The step of matching the comprehensive learning feature profile with a preset learning method template library to determine the best learning method that best matches the user's current knowledge source description includes: The knowledge source description is subjected to deep semantic parsing to generate a corresponding lightweight knowledge graph, which includes core concept nodes, logical relationships between concepts, and attribute tags of the knowledge domain. The comprehensive learning feature profile is aligned across dimensions with the lightweight knowledge graph to determine the fit score between the user's cognitive pattern and the knowledge structure features. Based on the fit score, the optimal matching item is retrieved from the preset learning method template library to determine the best learning method.

4. The control method for the learning machine as described in claim 1, characterized in that, The step of generating the corresponding learning task sequence according to the optimal learning method includes: Invoke the knowledge structuring transformation rules corresponding to the optimal learning method; Transform the knowledge content corresponding to the knowledge source description according to the knowledge structuring transformation rules to generate a knowledge node graph with logical hierarchy. Based on the user's learning habits, the knowledge node graph is divided into multiple learning task blocks, each of which has independent logical coherence. Generate an executable learning task and feedback node corresponding to each learning task block, wherein the feedback node includes embedded multiple-choice questions, knowledge repetition trigger points, or scenario simulation operations. The executable learning tasks are sorted by priority to generate the learning task sequence.

5. The control method for the learning machine as described in claim 1, characterized in that, The step of responding to an input operation for a knowledge source description and obtaining historical learning feature records corresponding to the knowledge source description includes: In response to the knowledge source description input by the user terminal, the key keywords and subject classification of the knowledge source description are parsed to determine the subject area and knowledge dimension to which the knowledge source belongs; Based on the subject area and knowledge dimension, retrieve historical learning feature records associated with the knowledge source from the local cache; If no matching record is found, establish a communication connection with the cloud server; Obtain historical learning feature records from the cloud server that match the knowledge source description.

6. The control method for the learning machine as described in claim 1, characterized in that, The step of adjusting the priority and content difficulty of the learning task sequence based on the execution feedback, optimizing the user's learning path, and updating the historical learning feature records includes: Analyze the execution feedback to identify the user's knowledge gaps, comprehension speed anomalies, and interest decay points in the learning task sequence; Based on the knowledge gaps and the comprehension speed anomalies, the content difficulty adjustment amount of the corresponding task nodes in the learning task sequence is determined. The content difficulty adjustment amount includes adding preparatory information, changing the explanation perspective, or breaking down complex steps. Based on the interest decay point, the priority of tasks that have not yet been executed in the learning task sequence is re-evaluated, and a corresponding task priority adjustment strategy is generated. Based on the content difficulty adjustment amount and task priority adjustment strategy, the learning task sequence is reconstructed to optimize the user's learning path; The optimized learning path is then updated in the historical learning feature record.

7. The control method for the learning machine as described in claim 1, characterized in that, The learning machine includes a display component; The process of displaying the learning task sequence and obtaining user feedback on the execution of the learning task sequence includes: Control the display component to render and display the learning task sequence in the form of a process graph or list; In response to a selection operation for any task node in the learning task sequence, the learning content corresponding to that task node is retrieved. In response to interactive operations on the learning content, time-series data of user behavior in content browsing, pausing, rewinding, accelerating, and answering questions are collected; The execution feedback is generated based on the time-series data, and the execution feedback includes task completion rate, interaction frequency, and abnormal intervals of dwell time.

8. The control method for the learning machine as described in claim 1, characterized in that, The step of obtaining the user's identification information and determining the user's corresponding learning habits includes: The user's unique identification information is obtained through the user's login credentials, and the user profile database in the cloud is accessed. Retrieve the user's long-term historical learning summary report from the cloud-based user profile database. The long-term historical learning summary report includes the user's click-through rate, completion rate, and active search preferences for learning resources in different media formats. Based on the long-term historical learning summary report, a multi-dimensional learning preference profile of the user is constructed to identify their preference for text, video or interactive content, and the learning habits are generated.

9. A control device, characterized in that, The control device includes: a memory, a processor, and a control program for the learning machine stored in the memory and executable on the processor, the control program for the learning machine being configured to implement the control method for the learning machine as described in any one of claims 1 to 8.

10. A learning machine, characterized in that, The learning machine includes: A display component, the display component being used to present a learning task sequence and an interactive interface; and The control device as claimed in claim 9, wherein the control device is electrically connected to the display component; The control device is used to acquire the user's identification information, determine the user's corresponding learning habits, and in response to the input operation for the knowledge source description, acquire the historical learning feature record corresponding to the knowledge source description. Then, based on the learning habits and the historical learning feature record, it generates the user's optimal learning method, generates the corresponding learning task sequence according to the optimal learning method, displays the learning task sequence, acquires the user's execution feedback on the learning task sequence, adjusts the priority and content difficulty of the learning task sequence based on the execution feedback, optimizes the user's learning path, and updates the historical learning feature record.

Images