Touch and talk pen interactive content generation method based on adaptive learning

Abstract

The invention discloses a touch and talk pen interactive content generation method based on adaptive learning. The method comprises the following steps: S1, acquiring a learning behavior data set of a user; s2, based on the learning behavior data set, constructing a dynamic knowledge graph of the user through an adaptive learning algorithm; s3, analyzing the learning behavior data set by using a query committee algorithm; s4, triggering a content generation module according to a bifurcation calculation result, wherein the content generation module generates learning content adaptive to the learning demand of the user based on the dynamic knowledge graph and the learning behavior data set of the user; s5, performing multi-mode interactive presentation of the learning content through the touch and talk pen; s6, performing feedback based on a learning behavior data set of the user for the learning content; s7, outputting a learning feedback report, wherein the learning feedback report is analyzed and generated based on the learning behavior data set of the user; and S8, repeating the steps S1 to S7. The interactive experience of the user is remarkably improved, the touch and talk pen is converted into an intelligent learning assistant from a passive content presentation tool, and efficient learning support and feedback are provided for the user.

Description

Technical Field

[0001] The present invention relates to the technical field of point reading pens, and in particular to a method for generating interactive content of a point reading pen based on adaptive learning. Background Art

[0002] With the rapid development of artificial intelligence technology, a series of intelligent learning tools have gradually emerged in the field of education to assist students in learning and mastering knowledge. As an educational tool that integrates voice, touch, and display multimodal interaction technologies, the reading pen is widely used in language learning and early childhood education. However, the content design and function implementation of traditional reading pens still have many limitations and cannot fully meet the needs of modern education for personalized, dynamic and intelligent learning.

[0003] At present, the reading pens on the market mainly rely on preset content for operation. The working principle is to trigger the corresponding audio, text or simple image in the built-in content library by scanning or identifying the marking information on the book. Although this design can provide basic voice broadcast and content display functions, it exposes the following problems in actual applications: First, the static nature of the content. The existing reading pen content library is a preset fixed resource and cannot be dynamically adjusted according to the user's learning behavior, knowledge mastery or interests, resulting in a monotonous learning experience and difficulty in meeting the personalized learning needs of different users. Second, the single interaction method. The interaction mode of the existing reading pen is mainly one-way touch or voice broadcast, lacking the deep integration of voice and touch and multimodal feedback mechanism, making it difficult to achieve intelligent interaction with users. Third, there is insufficient feedback on learning behavior. Traditional reading pens only provide content display functions and cannot collect and analyze users' learning data. They also lack in-depth analysis and learning path optimization based on learning data.

[0004] To sum up, the existing talking pen technology has significant deficiencies in the dynamics of content generation, in-depth analysis of user learning behavior, and flexibility of interaction methods. It is difficult to meet the needs of modern educational scenarios for intelligent and personalized learning. The defects of existing technology directly restrict the widespread application of talking pens in educational informatization. A new method is urgently needed to solve the above problems. Summary of the invention

[0005] One purpose of the present invention is to propose a method for generating interactive content of a reading pen based on adaptive learning. The present invention significantly improves the user's interactive experience, transforms the reading pen from a passive content presentation tool into an intelligent learning assistant, and provides users with efficient learning support and feedback.

[0006] A method for generating interactive content of a reading pen based on adaptive learning according to an embodiment of the present invention comprises the following steps:

[0007] S1. Obtain the user's learning behavior dataset;

[0008] S2. Based on the learning behavior dataset, a dynamic knowledge graph of users is constructed through an adaptive learning algorithm;

[0009] S3. Use the query committee algorithm to analyze the learning behavior data set, perform divergence calculation on the decision results generated by multiple learning models in the query committee algorithm, and determine whether the recommended content needs to be adjusted based on the divergence calculation results;

[0010] S4. triggering a content generation module according to the divergence calculation result, and generating learning content adapted to the user's learning needs based on the user's dynamic knowledge graph and learning behavior data set;

[0011] S5. Multimodal interactive presentation of learning content through the reading pen;

[0012] S6. Based on the user's feedback on the learning behavior dataset of the learning content, update the optimized learning path in the adaptive learning algorithm and the query committee algorithm in real time, adjust the relationship between knowledge points, and generate new learning paths;

[0013] S7. Output a learning feedback report, which is generated based on the analysis of the user's learning behavior data set, including learning progress, knowledge mastery, and recommended learning paths;

[0014] S8. Repeat steps S1 to S7 to form a continuous dynamic interaction cycle between the user and the reading pen.

[0015] Optionally, the S1 specifically includes:

[0016] S11. The touch module of the reading pen obtains the knowledge point i clicked by the user, and the knowledge point includes written text, picture annotations, and user-specified interactive targets;

[0017] S12. Record the click frequency f of users clicking on knowledge points i ,Click frequency indicates the number of times a user clicks on a specific knowledge point per unit time;

[0018] S13. Monitor the user's stay time T on the knowledge point i , obtained by calculating the click interval of the touch module;

[0019] S14. Obtaining learning feedback from the user, including correct and incorrect feedback annotations submitted by the user during the learning process through the reading pen touch module or the voice interaction module;

[0020] S15. Integrate the knowledge points clicked by the user, click frequency, dwell time and learning feedback data to construct a learning behavior data set D:

[0021] D={(i,f i ,T i ,F i )|i∈K};

[0022] Among them, F i is the learning feedback annotation of knowledge point i, and K is the set of all knowledge points.

[0023] Optionally, the S2 specifically includes:

[0024] S21. Extract the user's learning data on knowledge points from the learning behavior dataset D and calculate the weighted behavior score S for each knowledge point i i , the weighted behavior score represents the influence of learning behavior data on knowledge point i:

[0025]

[0026] Among them, S i is the weighted behavior score of the user on knowledge point i, log(1+f i ) is used to avoid numerical expansion when the frequency is too high. The denominator (1+γ·exp(-T i )) simulates the marginal decrease of learning effect after users stay for a long time, γ is the learning time adjustment parameter, sin(π·F i ) is used to highlight the different perceptions of users on correct and incorrect feedback. α, β, and δ are the weight coefficients of click frequency, dwell time, and learning feedback, respectively, indicating their relative importance in the calculation of behavior scores. max(F k ) is the maximum value of learning feedback among all knowledge points k∈K, which is used to normalize the influence of feedback annotation;

[0027] S22. Based on weighted behavior score S i Define the user's mastery of knowledge points function M(i):

[0028]

[0029] Among them, M(i) is the mastery level of knowledge point i, and the denominator is Represents the sum of the user's behavior scores on all knowledge points;

[0030] S23. Calculate the correlation R(i,j) between knowledge points based on the common attributes of knowledge points i and j in the dynamic knowledge graph;

[0031] S24. Based on the user's mastery of knowledge points M(i) and the correlation between knowledge points R(i,j), a dynamic knowledge graph is constructed:

[0032] G = (K, E);

[0033] Among them, E is the set of associated edges in the dynamic knowledge graph:

[0034] E={(i,j,W ij )|i,j∈K,R(i,j)·(M(i)+M(j))≥∈};

[0035] Among them, (M(i)+M(j)) is used to reflect the user's attention to the associated edge, W ij =R(i,j)·(M(i)+M(j)) represents the weight of the associated edge, which comprehensively measures the relevance of the knowledge point content and the user's learning status. ∈ is the set minimum effective weight threshold, which is used to filter low-weight edges;

[0036] S25. Update the dynamic knowledge graph G in real time, add new knowledge points or remove irrelevant knowledge points by adjusting the mastery level M(i) and relevance R(i,j) of knowledge points in real time.

[0037] Optionally, the S3 specifically includes:

[0038] S31. Initialize the committee model set C of the query committee algorithm = {M1, M2, ..., M k}, each committee model M j ∈C is constructed by an independent learning algorithm and trained based on the learning behavior dataset D;

[0039] S32. For each knowledge point in the learning behavior dataset, all committee models independently generate recommended content decisions

[0040]

[0041] in, For the committee model M j Recommended content for knowledge point i, g j Represents the committee model M j The decision function, Θ j For the committee model M j Parameter set;

[0042] S33. Calculate the divergence D(i) of the committee model for knowledge point i:

[0043]

[0044] Where D(i) is the divergence of knowledge point i, k is the number of committee models, is the indicator function, when the committee model M j and M lThe value is 1 when the decision results are inconsistent, otherwise it is 0;

[0045] S34. Determine whether the recommended content of knowledge point i needs to be marked or dynamically generated according to the divergence degree D(i):

[0046] If the divergence D(i)<δ, where δ is the set threshold, the current recommended content is retained;

[0047] If the divergence D(i) ≥ δ, the content annotation module or dynamic content generation module is triggered;

[0048] S35. For the knowledge point i that triggers annotation or is dynamically generated, update the decision model set and adjust the model parameters Θ through user feedback j ;

[0049] S36. Calculate the final recommended content based on the committee model’s divergence D(i) and the dynamic weight of the user’s learning behavior data

[0050]

[0051] Among them, the weight w j Combining the performance of the committee model and the dynamic adjustment of the reading pen scenario, the dynamic impact of user click frequency and dwell time in the committee model voting mechanism is reflected:

[0052]

[0053] Among them, w j For the committee model M j The weight of λ is the click frequency f i The weight adjustment factor, μ is the residence time T i The weight adjustment factor, σ j Represents the committee model M j The historical error rate.

[0054] Optionally, the S4 specifically includes:

[0055] S41. Identify the knowledge point set K that needs to generate content based on the divergence calculation results, the final recommended content, and the dynamic knowledge graph gen :

[0056]

[0057] Among them, K gen is the set of knowledge points that need to trigger content generation, R(i,j) is the correlation between knowledge points i and j;

[0058] S42. For each knowledge point i∈K genCombine the learning behavior dataset, dynamic knowledge graph and final recommended content to generate a content feature vector C that is adapted to the user's learning needs. i :

[0059]

[0060] Among them, C i is the content feature vector of knowledge point i, α1, α2, …, α6 are weight factors used to adjust the impact of click frequency, dwell time, learning feedback, mastery level, relevance and recommended content on content generation.

[0061] S43. Content feature vector C i Input to the content generation module, through the multimodal generation algorithm h(C i ) Dynamically generate learning content containing text, audio, video and image information:

[0062]

[0063] Among them, L i is the multimodal learning content of knowledge point i, Respectively represent the text, audio, video and image content of knowledge point i;

[0064] S44. Based on the generated multimodal learning content L i Verify the adaptability of user learning needs and calculate the adaptability function P(i);

[0065] S45. If the fitness function P(i)<η, η is the fitness threshold, the output of the content generation module is iteratively optimized, the weight factors α1, α2, ..., α6 are adjusted and the learning content is regenerated.

[0066] Optionally, the multimodal interactive presentation specifically includes:

[0067] The voice interaction module of the reading pen can realize content reading, extended information explanation and real-time response to user voice feedback;

[0068] The screen display module of the reading pen presents pictures, videos or dynamic knowledge graphs related to the learning content;

[0069] The touch control module of the reading pen enables users to click and select learning content, adjust the path, and mark points of interest.

[0070] Optionally, the S7 specifically includes:

[0071] S71. Based on the learning behavior data set D, the behavior characteristics of the user related to each knowledge point in the learning process are extracted to calculate the user's learning progress P u :

[0072]

[0073] Among them, P u For the user's learning progress, is the multimodal learning content of knowledge point i, represents the weight of content type m, F max Indicates the maximum value of learning feedback annotation;

[0074] S72. Based on dynamic knowledge graph, learning behavior dataset and multimodal learning content L i Generate a visual distribution of knowledge mastery, including the knowledge point mastery set M = {M(i)|i∈K} and the multimodal content coverage distribution corresponding to the knowledge point:

[0075]

[0076] in, represents the multimodal content coverage ratio of knowledge point i, is an indicator function, which takes the value of 1 when the corresponding modal content exists, otherwise it is 0;

[0077] S73. Combine the user's knowledge mastery level M(i) and the multimodal content coverage ratio The correlation degree R(i,j) of the knowledge points in the dynamic knowledge graph is used to recommend the next learning path set P next :

[0078]

[0079] S74. Generate a learning feedback report, which includes the user's learning progress P u and learning behavior characteristics, a visualization map M of user knowledge mastery and multimodal content coverage distribution The recommended next learning path set P next ;

[0080] S75. The learning feedback report is presented through the screen display module of the reading pen, and the voice interaction module is used to interpret the user's voice.

[0081] The beneficial effects of the present invention are:

[0082] (1) The present invention realizes real-time analysis of user learning status and accurate assessment of knowledge point mastery by constructing a dynamic knowledge graph based on user learning behavior data sets. The dynamic knowledge graph automatically optimizes the user's learning path and content recommendation by integrating learning behavior characteristics and the correlation between knowledge points. Compared with the fixed content library of traditional reading pens, it can dynamically generate personalized learning content that adapts to user learning needs, including text, audio, video and image multimodal content. The dynamic content recommendation based on the user's knowledge point mastery and correlation effectively improves the adaptability of learning content to user needs, solving the problem that traditional technologies cannot meet users' real-time learning needs.

[0083] (2) The present invention adopts a query committee algorithm to analyze the learning behavior data and constructs a committee composed of multiple independent models to vote on the recommended content, and triggers the content generation or labeling mechanism based on the divergence calculation results, which significantly improves the accuracy and robustness of the recommended content. It introduces a dynamic weight adjustment mechanism, combines the user's click frequency, residence time and learning feedback labeling, and dynamically optimizes the model's voting weight in the committee so that the recommended content can more accurately reflect the user's learning status and interests. Through multi-model collaboration and divergence calculation, the uncertainty of content recommendation is significantly reduced, ensuring the rationality and scientificity of the generated content.

[0084] (3) The present invention generates a learning feedback report including learning progress, knowledge point mastery and multimodal content coverage by integrating the user's learning behavior data and dynamic knowledge graph. The learning feedback report is presented in a multimodal form, providing visual learning analysis and intelligent recommendation of the next learning path. The validity of the generated content is verified by the fitness function and the content generation model is optimized in combination with user feedback, forming a closed-loop dynamic interaction mechanism, which significantly improves the user's interactive experience and transforms the reading pen from a passive content presentation tool to an intelligent learning assistant, providing users with efficient learning support and feedback. BRIEF DESCRIPTION OF THE DRAWINGS

[0085] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0086] Figure 1 The present invention provides a flowchart of a method for generating interactive content of a reading pen based on adaptive learning. DETAILED DESCRIPTION

[0087] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0088] refer to Figure 1 , a method for generating interactive content of a point reading pen based on adaptive learning, comprising the following steps:

[0089] S1. Obtain the user's learning behavior dataset;

[0090] S2. Based on the learning behavior dataset, a dynamic knowledge graph of users is constructed through an adaptive learning algorithm;

[0091] S3. Use the query committee algorithm to analyze the learning behavior data set, perform divergence calculation on the decision results generated by multiple learning models in the query committee algorithm, and determine whether the recommended content needs to be adjusted based on the divergence calculation results;

[0092] S4. triggering a content generation module according to the divergence calculation result, and generating learning content adapted to the user's learning needs based on the user's dynamic knowledge graph and learning behavior data set;

[0093] S5. Multimodal interactive presentation of learning content through the reading pen;

[0094] S6. Based on the user's feedback on the learning behavior dataset of the learning content, update the optimized learning path in the adaptive learning algorithm and the query committee algorithm in real time, adjust the relationship between knowledge points, and generate new learning paths;

[0095] S7. Output a learning feedback report, which is generated based on the analysis of the user's learning behavior data set, including learning progress, knowledge mastery, and recommended learning paths;

[0096] S8. Repeat steps S1 to S7 to form a continuous dynamic interaction cycle between the user and the reading pen.

[0097] In this implementation, S1 specifically includes:

[0098] S11. The touch module of the reading pen obtains the knowledge point i clicked by the user, and the knowledge point includes written text, picture annotations, and user-specified interactive targets;

[0099] S12. Record the click frequency f of users clicking on knowledge points i ,Click frequency indicates the number of times a user clicks on a specific knowledge point per unit time;

[0100] S13. Monitor the user's stay time T on the knowledge point i , obtained by calculating the click interval of the touch module;

[0101] S14. Obtaining learning feedback from the user, including correct and incorrect feedback annotations submitted by the user during the learning process through the reading pen touch module or the voice interaction module;

[0102] S15. Integrate the knowledge points clicked by users, click frequency, dwell time and learning feedback data to construct a learning behavior dataset D:

[0103] D={(i,f i ,T i ,F i )|i∈K};

[0104] Among them, F i is the learning feedback annotation of knowledge point i, and K is the set of all knowledge points.

[0105] In this implementation, S2 specifically includes:

[0106] S21. Extract the user's learning data on knowledge points from the learning behavior dataset D and calculate the weighted behavior score S for each knowledge point i i , the weighted behavior score represents the influence of learning behavior data on knowledge point i:

[0107]

[0108] Among them, S i is the weighted behavior score of the user on knowledge point i, log(1+f i ) is used to avoid numerical expansion when the frequency is too high. The denominator (1+γ·exp(-T i )) simulates the marginal decrease of learning effect after users stay for a long time, γ is the learning time adjustment parameter, sin(π·F i ) is used to highlight the different perceptions of users on correct and incorrect feedback. α, β, and δ are the weight coefficients of click frequency, dwell time, and learning feedback, respectively, indicating their relative importance in the calculation of behavior scores. max(F k ) is the maximum value of learning feedback among all knowledge points k∈K, which is used to normalize the influence of feedback annotation;

[0109] S22. Based on weighted behavior score S i Define the user's mastery of knowledge points function M(i):

[0110]

[0111] Among them, M(i) is the mastery level of knowledge point i, and the denominator ∑ j∈K S j Represents the sum of the user's behavior scores on all knowledge points;

[0112] S23. Calculate the correlation R(i,j) between knowledge points based on the common attributes of knowledge points i and j in the dynamic knowledge graph;

[0113] S24. Based on the user's mastery of knowledge points M(i) and the correlation between knowledge points R(i,j), a dynamic knowledge graph is constructed:

[0114] G = (K, E);

[0115] Among them, E is the set of associated edges in the dynamic knowledge graph:

[0116] E={(i,j,W ij )|i,j∈K,R(i,j)·(M(i)+M(j))≥∈};

[0117] Among them, (M(i)+M(j)) is used to reflect the user's attention to the associated edge, W ij =R(i,j)·(M(i)+M(j)) represents the weight of the associated edge, which comprehensively measures the relevance of the knowledge point content and the user's learning status. ∈ is the set minimum effective weight threshold, which is used to filter low-weight edges;

[0118] S25. Update the dynamic knowledge graph G in real time, add new knowledge points or remove irrelevant knowledge points by adjusting the mastery level M(i) and relevance R(i,j) of knowledge points in real time.

[0119] In this implementation, S3 specifically includes:

[0120] S31. Initialize the committee model set C of the query committee algorithm = {M1, M2, ..., M k}, each committee model M j ∈C is constructed by an independent learning algorithm and trained based on the learning behavior dataset D;

[0121] S32. For each knowledge point in the learning behavior dataset, all committee models independently generate recommended content decisions

[0122]

[0123] in, For the committee model M j Recommended content for knowledge point i, g j Represents the committee model M j The decision function, Θ j For the committee model M j Parameter set;

[0124] S33. Calculate the divergence D(i) of the committee model for knowledge point i:

[0125]

[0126] Where D(i) is the divergence of knowledge point i, k is the number of committee models, is the indicator function, when the committee model M j and M l The value is 1 when the decision results are inconsistent, otherwise it is 0;

[0127] S34. Determine whether the recommended content of knowledge point i needs to be marked or dynamically generated according to the divergence degree D(i):

[0128] If the divergence D(i)<δ, where δ is the set threshold, the current recommended content is retained;

[0129] If the divergence D(i) ≥ δ, the content annotation module or dynamic content generation module is triggered;

[0130] S35. For the knowledge point i that triggers annotation or is dynamically generated, update the decision model set and adjust the model parameters Θ through user feedback j ;

[0131] S36. Calculate the final recommended content based on the committee model’s divergence D(i) and the dynamic weight of the user’s learning behavior data

[0132]

[0133] Among them, the weight w j Combining the performance of the committee model and the dynamic adjustment of the reading pen scenario, the dynamic impact of user click frequency and dwell time in the committee model voting mechanism is reflected:

[0134]

[0135] Among them, w j For the committee model M j The weight of λ is the click frequency f i The weight adjustment factor, μ is the residence time T i The weight adjustment factor, σ j Represents the committee model M j The historical error rate.

[0136] In this implementation, S4 specifically includes:

[0137] S41. Identify the knowledge point set K that needs to generate content based on the divergence calculation results, the final recommended content, and the dynamic knowledge graph gen :

[0138]

[0139] Among them, K genis the set of knowledge points that need to trigger content generation, R(i,j) is the correlation between knowledge points i and j;

[0140] S42. For each knowledge point i∈K gen Combine the learning behavior dataset, dynamic knowledge graph and final recommended content to generate a content feature vector C that is adapted to the user's learning needs. i :

[0141]

[0142] Among them, C i is the content feature vector of knowledge point i, α1, α2, …, α6 are weight factors used to adjust the impact of click frequency, dwell time, learning feedback, mastery level, relevance and recommended content on content generation.

[0143] S43. Content feature vector C i Input to the content generation module, through the multimodal generation algorithm h(C i ) Dynamically generate learning content containing text, audio, video and image information:

[0144]

[0145] Among them, L i is the multimodal learning content of knowledge point i, Respectively represent the text, audio, video and image content of knowledge point i;

[0146] S44. Based on the generated multimodal learning content L i Verify the adaptability of user learning needs and calculate the adaptability function P(i);

[0147] S45. If the fitness function P(i)<η, η is the fitness threshold, the output of the content generation module is iteratively optimized, the weight factors α1, α2, ..., α6 are adjusted and the learning content is regenerated.

[0148] In this implementation, the multimodal interactive presentation specifically includes:

[0149] The voice interaction module of the reading pen can realize content reading, extended information explanation and real-time response to user voice feedback;

[0150] The screen display module of the reading pen presents pictures, videos or dynamic knowledge graphs related to the learning content;

[0151] The touch control module of the reading pen enables users to click and select learning content, adjust the path, and mark points of interest.

[0152] In this implementation, S7 specifically includes:

[0153] S71. Based on the learning behavior data set D, the behavior characteristics of the user related to each knowledge point in the learning process are extracted to calculate the user's learning progress P u :

[0154]

[0155] Among them, P u For the user's learning progress, is the multimodal learning content of knowledge point i, represents the weight of content type m, F max Indicates the maximum value of learning feedback annotation;

[0156] S72. Based on dynamic knowledge graph, learning behavior dataset and multimodal learning content L i Generate a visual distribution of knowledge mastery, including the knowledge point mastery set M = {M(i)|i∈K} and the multimodal content coverage distribution corresponding to the knowledge point:

[0157]

[0158] in, represents the multimodal content coverage ratio of knowledge point i, is an indicator function, which takes the value of 1 when the corresponding modal content exists, otherwise it is 0;

[0159] S73. Combine the user's knowledge mastery level M(i) and the multimodal content coverage ratio The correlation degree R(i,j) of the knowledge points in the dynamic knowledge graph is used to recommend the next learning path set P next :

[0160]

[0161] S74. Generate a learning feedback report, which includes the user's learning progress P u and learning behavior characteristics, a visualization map M of user knowledge mastery and multimodal content coverage distribution The recommended next learning path set P next ;

[0162] S75. The learning feedback report is presented through the screen display module of the reading pen, and the voice interaction module is used to interpret the user's voice.

[0163] Embodiment 1:

[0164] Embodiment In a third-grade classroom of an elementary school located in Area A, students are using a reading pen to assist in English learning. The theme of this class is "Description of Family Members" ("My mother is a teacher."). The teaching content of a traditional reading pen only includes simple word pronunciation and sentence reading. It does not have the ability to generate dynamic learning content and is difficult to adapt to students' different learning needs. In order to improve students' interest and efficiency in English learning, the teacher chooses to introduce the method of the present invention. Through the reading pen based on adaptive learning and query committee algorithm, it provides students with real-time generated personalized learning content, helps teachers comprehensively analyze students' learning status, and optimizes teaching strategies.

[0165] When students start learning English, each of them holds a reading pen that supports the method of the present invention. The reading pen is pre-installed with textbook texts and related image and audio resource libraries, but the actual presentation of the content is driven by a dynamic knowledge graph and learning behavior data. The goal of this study is to master 10 core words and 3 sentence patterns, while cultivating students' sentence pattern-making ability.

[0166] During use, the reading pen records students' learning behavior data through the touch module. When a student was learning "mother", he clicked the word three times, and stayed for 4 seconds, 6 seconds and 3 seconds each time. The reading pen calculated the learning behavior characteristics of this knowledge point based on the student's click frequency and dwell time, and constructed a learning behavior data set in combination with learning feedback annotations (whether the example sentence question was answered correctly). The dynamic knowledge graph is updated in real time to record the student's mastery of the word "mother" and its correlation with other knowledge points ("father", "teacher").

[0167] To generate personalized learning content, the reading pen uses the query committee algorithm to analyze students' learning data. The committee model makes independent decisions based on students' click behavior, mastery level, and error rate, and finally obtains recommended content through divergence calculation. For students' errors in "mother" related sentences, the reading pen generates multimodal extended content including text, audio, images, and videos, including:

[0168] Read the sentence "My mother is a teacher" aloud and guide students to imitate pronunciation with the help of lip-syncing video;

[0169] Provide pictures of family members and add the sentence pattern "My father is a doctor." to help students understand the logic of the sentence pattern;

[0170] Asking “What does your mother do?” guides students to construct similar sentence patterns.

[0171] At the end of the class, the reading pen presents each student's learning feedback report on the screen. The report includes learning progress (the number of words and sentence patterns mastered), knowledge point distribution (displayed in a radar chart) and the recommended next learning path. A student's report shows that his learning progress is 80%. The knowledge points with a high degree of mastery include "mother" and "teacher", but there are certain weaknesses in "brother" and "nurse". The recommended learning path is to review the weak points first and strengthen sentence-making exercises.

[0172] In order to verify the effectiveness of the method of the present invention, a two-week comparative experiment was conducted in two primary schools in District A. The experiment was divided into two groups: the experimental group used the point reading pen of the present invention, and the control group used the traditional point reading pen (fixed content library). Both groups studied with the same learning goals (10 words, 3 sentence patterns), and each group had 30 students. The data results are as follows:

[0173]

[0174] At the same time, the dynamic behavior data of two groups of reading pens during the learning process were recorded:

[0175] Learning content adaptation rate: The adaptation rate of the learning content generated by the experimental group's reading pen based on the dynamic knowledge graph is 93%, which is much higher than the adaptation rate of the fixed content of the control group's reading pen of 67%.

[0176] Learning path optimization rate: The experimental group's reading pen optimized the learning path 240 times in two weeks (8 times per person), while the control group was unable to perform path optimization.

[0177] Error review coverage: The experimental group's reading pen reviewed all incorrect knowledge points in the personalized learning content (coverage rate 100%), while the control group only covered about 58% of the incorrect knowledge points through mechanical repetition review.

[0178] It can be seen from the above embodiments that the method of the present invention has significant advantages in primary school English learning scenarios. Through the combination of dynamic knowledge graph and query committee algorithm, the reading pen can realize the dynamic and accurate content generation, while greatly improving students' learning efficiency and interest. Comparative experimental data further proves that the method of the present invention is superior to traditional methods in terms of learning progress, mastery accuracy, improvement of learning interest and time efficiency, and provides an innovative solution for educational informatization.

[0179] The present invention realizes real-time analysis of user learning status and accurate assessment of knowledge point mastery by constructing a dynamic knowledge graph based on user learning behavior data sets. The dynamic knowledge graph automatically optimizes the user's learning path and content recommendation by comprehensively analyzing learning behavior characteristics and the correlation between knowledge points. Compared with the fixed content library of traditional reading pens, it can dynamically generate personalized learning content that adapts to user learning needs, including text, audio, video and image multimodal content. The dynamic content recommendation based on the user's knowledge point mastery and correlation effectively improves the adaptability of learning content to user needs, solving the problem that traditional technology cannot meet users' real-time learning needs.

[0180] The present invention adopts a query committee algorithm to analyze the learning behavior data and constructs a committee composed of multiple independent models to vote on the recommended content, and triggers the content generation or annotation mechanism based on the divergence calculation results, which significantly improves the accuracy and robustness of the recommended content. It introduces a dynamic weight adjustment mechanism, combines the user's click frequency, residence time and learning feedback annotation, and dynamically optimizes the model's voting weight in the committee so that the recommended content can more accurately reflect the user's learning status and interests. Through multi-model collaboration and divergence calculation, the uncertainty of content recommendation is significantly reduced, ensuring the rationality and scientificity of the generated content.

[0181] The present invention generates a learning feedback report including learning progress, knowledge point mastery and multimodal content coverage by integrating the user's learning behavior data and dynamic knowledge graph. The learning feedback report is presented in a multimodal form, providing visual learning analysis and intelligent recommendation of the next learning path. The validity of the generated content is verified by the fitness function and the content generation model is optimized in combination with user feedback, forming a closed-loop dynamic interaction mechanism, which significantly improves the user's interactive experience, transforming the reading pen from a passive content presentation tool to an intelligent learning assistant, and providing users with efficient learning support and feedback.

[0182] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for generating interactive content of a point reading pen based on adaptive learning, characterized in that: The steps include: S1. Obtain the user's learning behavior dataset; S2. Based on the learning behavior dataset, a dynamic knowledge graph of users is constructed through an adaptive learning algorithm; S3. Use the query committee algorithm to analyze the learning behavior data set, perform divergence calculation on the decision results generated by multiple learning models in the query committee algorithm, and determine whether the recommended content needs to be adjusted based on the divergence calculation results; S4. triggering a content generation module according to the divergence calculation result, and generating learning content adapted to the user's learning needs based on the user's dynamic knowledge graph and learning behavior data set; S5. Multimodal interactive presentation of learning content through the reading pen; S6. Based on the user's feedback on the learning behavior dataset of the learning content, update the optimized learning path in the adaptive learning algorithm and the query committee algorithm in real time, adjust the relationship between knowledge points, and generate new learning paths; S7. Output a learning feedback report, which is generated based on the analysis of the user's learning behavior data set, including learning progress, knowledge mastery, and recommended learning paths; S8. Repeat steps S1 to S7 to form a continuous dynamic interaction cycle between the user and the reading pen.

2. The method for generating interactive content of a point reading pen based on adaptive learning according to claim 1, characterized in that: The S1 specifically includes: S11. The touch module of the reading pen obtains the knowledge point i clicked by the user, and the knowledge point includes written text, picture annotations, and user-specified interactive targets; S12. Record the click frequency f of users clicking on knowledge points i ,Click frequency indicates the number of times a user clicks on a specific knowledge point per unit time; S13. Monitor the user's stay time T on the knowledge point i , obtained by calculating the click interval of the touch module; S14. Obtaining learning feedback from the user, including correct and incorrect feedback annotations submitted by the user during the learning process through the reading pen touch module or the voice interaction module; S15. Integrate the knowledge points clicked by the user, click frequency, dwell time and learning feedback data to construct a learning behavior data set D: D={(i,f i ,T i ,F i )|i∈K}; Among them, F i is the learning feedback annotation of knowledge point i, and K is the set of all knowledge points.

3. The method for generating interactive content of a point reading pen based on adaptive learning according to claim 2, characterized in that: The S2 specifically includes: S21. Extract the user's learning data on knowledge points from the learning behavior dataset D and calculate the weighted behavior score S for each knowledge point i i , the weighted behavior score represents the influence of learning behavior data on knowledge point i: Among them, S i is the weighted behavior score of the user on knowledge point i, log(1+f i ) is used to avoid numerical expansion when the frequency is too high. The denominator (1+γ·exp(-T i )) simulates the marginal decrease of learning effect after users stay for a long time, γ is the learning time adjustment parameter, sin(π·F i ) is used to highlight the different perceptions of users on correct and incorrect feedback. α, β, and δ are the weight coefficients of click frequency, dwell time, and learning feedback, respectively, indicating their relative importance in the calculation of behavior scores. max(F k ) is the maximum value of learning feedback among all knowledge points k∈K, which is used to normalize the influence of feedback annotation; S22. Based on weighted behavior score S i Define the user's mastery of knowledge points function M(i): Among them, M(i) is the mastery level of knowledge point i, and the denominator ∑ j∈K S j Represents the sum of the user's behavior scores on all knowledge points; S23. Calculate the correlation R(i,j) between knowledge points based on the common attributes of knowledge points i and j in the dynamic knowledge graph; S24. Based on the user's mastery of knowledge points M(i) and the correlation between knowledge points R(i,j), a dynamic knowledge graph is constructed: G = (K, E); Among them, E is the set of associated edges in the dynamic knowledge graph: E={(i,j,W ij )|i,j∈K,R(i,j)·(M(i)+M(j))≥∈}; Among them, (M(i)+M(j)) is used to reflect the user's attention to the associated edge, W ij =R(i,j)·(M(i)+M(j)) represents the weight of the associated edge, which comprehensively measures the relevance of the knowledge point content and the user's learning status. ∈ is the set minimum effective weight threshold, which is used to filter low-weight edges; S25. Update the dynamic knowledge graph G in real time, add new knowledge points or remove irrelevant knowledge points by adjusting the mastery level M(i) and relevance R(i,j) of knowledge points in real time.

4. The method for generating interactive content of a point reading pen based on adaptive learning according to claim 1, characterized in that: The S3 specifically includes: S31. Initialize the committee model set C of the query committee algorithm = {M1, M2, ..., M k }, each committee model M j ∈C is constructed by an independent learning algorithm and trained based on the learning behavior dataset D; S32. For each knowledge point in the learning behavior dataset, all committee models independently generate recommended content decisions in, For the committee model M j Recommended content for knowledge point i, g j Represents the committee model M j The decision function, Θ j For the committee model M j Parameter set; S33. Calculate the divergence D(i) of the committee model for knowledge point i: Where D(i) is the divergence of knowledge point i, k is the number of committee models, is the indicator function, when the committee model M j and M l The value is 1 when the decision results are inconsistent, otherwise it is 0; S34. Determine whether the recommended content of knowledge point i needs to be marked or dynamically generated according to the divergence degree D(i): If the divergence D(i)<δ, where δ is the set threshold, the current recommended content is retained; If the divergence D(i) ≥ δ, the content annotation module or dynamic content generation module is triggered; S35. For the knowledge point i that triggers annotation or is dynamically generated, update the decision model set and adjust the model parameters Θ through user feedback j ; S36. Calculate the final recommended content based on the committee model’s divergence D(i) and the dynamic weight of the user’s learning behavior data Among them, the weight w j Combining the performance of the committee model and the dynamic adjustment of the reading pen scenario, the dynamic impact of user click frequency and dwell time in the committee model voting mechanism is reflected: Among them, w j For the committee model M j The weight of λ is the click frequency f i The weight adjustment factor, μ is the residence time T i The weight adjustment factor, σ j Represents the committee model M j The historical error rate.

5. The method for generating interactive content of a point reading pen based on adaptive learning according to claim 4, characterized in that: The S4 specifically includes: S41. Identify the knowledge point set K that needs to generate content based on the divergence calculation results, the final recommended content, and the dynamic knowledge graph gen : Among them, K gen is the set of knowledge points that need to trigger content generation, R(i,j) is the correlation between knowledge points i and j; S42. For each knowledge point i∈K gen Combine the learning behavior dataset, dynamic knowledge graph and final recommended content to generate a content feature vector C that is adapted to the user's learning needs. i : Among them, C i is the content feature vector of knowledge point i, α1, α2, …, α6 are weight factors used to adjust the impact of click frequency, dwell time, learning feedback, mastery level, relevance and recommended content on content generation. S43. The content feature vector C i Input to the content generation module, through the multimodal generation algorithm h(C i ) Dynamically generate learning content containing text, audio, video and image information: Among them, L i is the multimodal learning content of knowledge point i, Respectively represent the text, audio, video and image content of knowledge point i; S44. Based on the generated multimodal learning content L i Verify the adaptability of user learning needs and calculate the adaptability function P(i); S45. If the fitness function P(i)<η, η is the fitness threshold, the output of the content generation module is iteratively optimized, the weight factors α1, α2, ..., α6 are adjusted and the learning content is regenerated.

6. The method for generating interactive content of a point reading pen based on adaptive learning according to claim 1, characterized in that: The multimodal interactive presentation specifically includes: The voice interaction module of the reading pen can realize content reading, extended information explanation and real-time response to user voice feedback; The screen display module of the reading pen presents pictures, videos or dynamic knowledge graphs related to the learning content; The touch control module of the reading pen enables users to click and select learning content, adjust the path, and mark points of interest.

7. The method for generating interactive content of a point reading pen based on adaptive learning according to claim 1, characterized in that: The S7 specifically includes: S71. Based on the learning behavior data set D, the behavior characteristics of the user related to each knowledge point in the learning process are extracted to calculate the user's learning progress P u : Among them, P u For the user's learning progress, is the multimodal learning content of knowledge point i, represents the weight of content type m, F max Indicates the maximum value of learning feedback annotation; S72. Based on dynamic knowledge graph, learning behavior dataset and multimodal learning content L i Generate a visual distribution of knowledge mastery, including the knowledge point mastery set M = {M(i)|i∈K} and the multimodal content coverage distribution corresponding to the knowledge point: in, represents the multimodal content coverage ratio of knowledge point i, is an indicator function, which takes the value of 1 when the corresponding modal content exists, otherwise it is 0; S73. Combine the user's knowledge mastery level M(i) and the multimodal content coverage ratio The correlation degree R(i,j) of the knowledge points in the dynamic knowledge graph is used to recommend the next learning path set P next : S74. Generate a learning feedback report, which includes the user's learning progress P u and learning behavior characteristics, a visualization map M of user knowledge mastery and multimodal content coverage distribution The recommended next learning path set P next ; S75. The learning feedback report is presented through the screen display module of the reading pen, and the voice interaction module is used to interpret the user's voice.

Images