Intelligent learning method and device

Abstract

The application relates to the technical field of computers, in particular to an intelligent learning method and device, which comprises the following steps: collecting student portrait data and determining teaching target data; formulating a personalized learning scheme through a reinforcement learning model according to the student portrait data and the teaching target data; in the execution process of the learning scheme, abstract knowledge is converted into a visual and interactive three-dimensional model through virtual reality and augmented reality technology to assist students in learning, and a dialogue exchange platform is built to interactively answer questions; obtaining problem texts uploaded by students in the learning process, analyzing learning intentions from the problem texts, and extracting core elements in the problem texts; and outputting answers conforming to the cognitive level of students through a large model according to the student portrait data, the analyzed learning intentions and the extracted core elements. The application helps to improve learning efficiency.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to an intelligent learning companion method and device. Background Technology

[0002] Intelligent learning support is an innovative educational model that integrates advanced information technology and educational theory. Its core lies in providing students with all-weather, personalized, and precise learning support services through artificial intelligence technology, big data analysis, and adaptive learning algorithms.

[0003] In the current education field, with the deep integration of information technology and teaching, intelligent learning companionship is gradually becoming an important direction for promoting personalized education reform. While traditional intelligent learning companionship methods have initially applied big data and simple algorithms, they still have many limitations. For example, most solutions are based on a unified teaching syllabus to formulate learning plans, failing to fully consider individual differences in students' knowledge base, learning abilities, and interests, resulting in a "one-size-fits-all" approach that fails to meet the learning needs of different students. Secondly, existing Q&A systems often match answers based on pre-set question banks, and the output answers may exceed students' cognitive levels or be too simplistic. Furthermore, traditional learning companionship relies heavily on two-dimensional presentations such as text and images, making it difficult for students to intuitively understand the essence of knowledge. These limitations lead to poor learning experiences and lower levels of knowledge mastery, thus affecting learning outcomes. Summary of the Invention

[0004] Therefore, it is necessary to provide an intelligent learning companion method and device that can improve learning efficiency in response to the above-mentioned technical problems.

[0005] Firstly, this application provides an intelligent learning companion method, the method comprising:

[0006] Collect student profile data and define teaching objective data;

[0007] Based on the student profile data and the teaching objective data, a personalized learning plan is formulated using a reinforcement learning model. The learning plan includes the learning order of knowledge points and the corresponding learning time.

[0008] During the implementation of the learning program, virtual reality and augmented reality technologies are used to transform abstract knowledge into visual and interactive three-dimensional models to assist students in learning and to build a dialogue and communication platform for interactive Q&A.

[0009] The system obtains the text of questions uploaded by students during the learning process, parses out the learning intentions based on the question text, and extracts the core elements from the question text.

[0010] Based on the student profile data, the analyzed learning intentions, and the extracted core elements, a large model is used to output answers that match the students' cognitive levels.

[0011] In one embodiment, the step of outputting an answer that matches the student's cognitive level based on the student profile data, the parsed learning intentions, and the extracted core elements through a large model includes:

[0012] Based on the student profile data, the students' knowledge coverage, accuracy of understanding knowledge points, and ability to apply knowledge points are assessed to quantify their knowledge level.

[0013] Based on the knowledge level, the learning intention obtained from the analysis, and the extracted core elements, an initial answer that matches the student's cognitive level is obtained through a large model;

[0014] The initial answer is retrieved and validated using a pre-defined domain knowledge base to avoid errors. A pre-defined evaluation model is then used to filter out awkward expressions in the initial answer before outputting the final answer.

[0015] In one embodiment, the step of developing a personalized learning plan based on the student profile data and the teaching objective data using a reinforcement learning model includes:

[0016] Based on the teaching objective data, a knowledge point dependency graph is constructed, which includes the preceding relationship, parallel relationship and progressive relationship of knowledge points;

[0017] Based on the knowledge point dependency graph and the student profile data, a personalized learning plan is developed using a reinforcement learning model.

[0018] In one embodiment, the step of obtaining the question text uploaded by the student during the learning process, parsing the learning intention based on the question text, and extracting the core elements from the question text includes:

[0019] The problem text is standardized and encoded to generate an encoded feature representation of the current text;

[0020] Obtain the encoded feature representation of the historical dialogue, concatenate the encoded feature representation of the current text and the encoded feature representation of the historical dialogue, and capture the intent association through an attention mechanism to parse out the learning intent;

[0021] Core elements are extracted from the problem text using named entity recognition and semantic role labeling technologies.

[0022] In one embodiment, the intelligent learning companion method further includes:

[0023] For scarce disciplines, few-shot learning and prompting engineering are employed to activate the domain transfer capability of the large model through a small amount of question-and-answer data;

[0024] We collected spoken language data from students in different regions, constructed a dialect standard mapping table, and enhanced the tolerance of the large model to non-standard expressions through comparative learning.

[0025] In one embodiment, the intelligent learning companion method further includes:

[0026] By using sequence pattern mining algorithms, timestamp sequences of student behavior are analyzed to extract high-frequency learning behavior paths;

[0027] Based on the extracted high-frequency learning behavior paths, distinguish between efficient and inefficient learning modes;

[0028] Provide real-time reminders to students who are in an inefficient learning mode.

[0029] In one embodiment, after formulating a personalized learning plan based on the student profile data and the teaching objective data using a reinforcement learning model, the method further includes:

[0030] Real-time collection of student learning feedback data, including the accuracy rate of completing the learning, the time spent, and the satisfaction with proactive feedback;

[0031] The learning feedback data is input into the reinforcement learning model to update the learning scheme.

[0032] Secondly, this application also provides an intelligent learning companion device. The device includes:

[0033] The data acquisition module is used to collect student profile data and define teaching objective data;

[0034] The learning plan development module is used to develop personalized learning plans based on the student profile data and the teaching objective data through a reinforcement learning model. The learning plan includes the learning order of knowledge points and the corresponding learning duration.

[0035] The knowledge learning module is used to transform abstract knowledge into a visual and interactive 3D model through virtual reality and augmented reality technologies during the execution of the learning program, in order to assist students in learning and to build a dialogue and communication platform for interactive Q&A.

[0036] The question text parsing module is used to obtain the question text uploaded by students during the learning process, parse out the learning intention based on the question text, and extract the core elements in the question text;

[0037] The answer output module is used to output answers that match the students' cognitive level based on the student profile data, the parsed learning intentions, and the extracted core elements through a large model.

[0038] Thirdly, this application also provides a computer device. The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to perform the following steps:

[0039] Collect student profile data and define teaching objective data;

[0040] Based on the student profile data and the teaching objective data, a personalized learning plan is formulated using a reinforcement learning model. The learning plan includes the learning order of knowledge points and the corresponding learning time.

[0041] During the implementation of the learning program, virtual reality and augmented reality technologies are used to transform abstract knowledge into visual and interactive three-dimensional models to assist students in learning and to build a dialogue and communication platform for interactive Q&A.

[0042] The system obtains the text of questions uploaded by students during the learning process, parses out the learning intentions based on the question text, and extracts the core elements from the question text.

[0043] Based on the student profile data, the analyzed learning intentions, and the extracted core elements, a large model is used to output answers that match the students' cognitive levels.

[0044] Fourthly, this application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, performs the following steps:

[0045] Collect student profile data and define teaching objective data;

[0046] Based on the student profile data and the teaching objective data, a personalized learning plan is formulated using a reinforcement learning model. The learning plan includes the learning order of knowledge points and the corresponding learning time.

[0047] During the implementation of the learning program, virtual reality and augmented reality technologies are used to transform abstract knowledge into visual and interactive three-dimensional models to assist students in learning and to build a dialogue and communication platform for interactive Q&A.

[0048] The system obtains the text of questions uploaded by students during the learning process, parses out the learning intentions based on the question text, and extracts the core elements from the question text.

[0049] Based on the student profile data, the analyzed learning intentions, and the extracted core elements, a large model is used to output answers that match the students' cognitive levels.

[0050] In summary, this application includes the following beneficial technical effects:

[0051] By combining student profile data with reinforcement learning models, learning programs can be precisely matched to students' knowledge base, abilities, and preferences, avoiding a one-size-fits-all approach and ensuring that students of different levels receive appropriate learning content. Through virtual reality and augmented reality technologies, abstract knowledge is transformed into visual, interactive 3D models, enabling students to understand the essence of knowledge through immersive experiences. By analyzing learning intentions and extracting core elements, answers that match students' cognitive levels are output, avoiding irrelevant or mismatched answers and improving the effectiveness of Q&A. This ensures that the learning process always aligns with students' real-time needs and maximizes learning outcomes. Attached Figure Description

[0052] Figure 1 This is a flowchart illustrating an intelligent learning companion method in one embodiment;

[0053] Figure 2 This is a flowchart illustrating the intelligent learning companion method in another embodiment;

[0054] Figure 3 This is a structural block diagram of an intelligent learning companion device in one embodiment. Detailed Implementation

[0055] This invention provides an intelligent learning companion method and device.

[0056] The embodiments of the present invention will now be described in more detail with reference to the accompanying drawings. While some embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the present disclosure. It should be understood that the accompanying drawings and embodiments are for illustrative purposes only and are not intended to limit the scope of protection of the present invention.

[0057] In the description of the embodiments disclosed in this invention, the term "comprising" and similar terms should be understood as open-ended inclusion, i.e., "including but not limited to". The term "based on" should be understood as "at least partially based on". The term "one embodiment" or "the embodiment" should be understood as "at least one embodiment". The terms "first", "second", etc., may refer to different or the same objects. Other explicit and implicit definitions may also be included below.

[0058] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 One embodiment of the intelligent learning companion method in this invention includes:

[0059] S100: Collect student profile data and define teaching objective data.

[0060] Specifically, student profile data forms the basis for developing personalized learning plans, while teaching objective data provides guidance for learning direction. This student profile data includes students' knowledge base, learning ability, and behavioral preference data. Knowledge base data includes students' mastery of knowledge points in each subject and the distribution of knowledge gaps; this data is obtained through analysis of entrance tests, daily assignments, and past exam scores. Learning ability data records students' learning time, logical reasoning ability, and memory ability. Behavioral preference data includes students' preferred learning time periods, learning methods, and areas of interest. Teaching objective data needs to be determined in conjunction with educational standards, the needs of the learning stage, and individual student goals. This includes macro-level goals, micro-level goals, and personalized goals. Macro-level goals are the core tasks for a specific learning stage / cycle, determined based on national curriculum standards or industry training specifications. Micro-level goals are individual knowledge points or learning units broken down from macro-level goals. Personalized goals are customized goals supplemented by individual student needs; for example, a student might want to improve their English listening comprehension to 80% of the maximum score within one month.

[0061] S200 uses a reinforcement learning model to develop personalized learning plans based on student profile data and teaching objective data.

[0062] The learning plan includes the learning order of knowledge points and the corresponding learning time.

[0063] Specifically, a reinforcement learning model is first constructed, which defines a state space, an action space, and a reward function. The state space takes the student's current knowledge state and the progress of the teaching objectives as the core states. For example, if the student's current mastery of the quadratic equation knowledge point is 60% (basic formula mastery, weak application), the teaching objective requires a mastery of 90% for this knowledge point, while also connecting to the subsequent quadratic function knowledge point. The action space contains adjustable parameters of the learning plan, specifically the learning order of knowledge points and the allocation of learning time. The reward function calculates the reward value based on the improvement of knowledge mastery, learning efficiency, and interest matching degree as core indicators. The reinforcement learning model first categorizes students into "weak foundation," "advanced improvement," and "advanced extension" types based on student profile data. Then, it breaks down the knowledge point system by combining teaching objective data to generate a suitable learning plan. For example, for "weak foundation" students (who have a vague memory of the quadratic equation formula and prefer short video learning), the plan is as follows: first, review the definition and formula of the quadratic equation through a 15-minute short video; then, use 45 minutes of VR to demonstrate the derivation of the formula; next, complete basic formula application exercises in 1 hour (with real-time error correction); and finally, summarize common mistakes in 30 minutes.

[0064] During the implementation of the learning program, the S300 uses virtual reality and augmented reality technologies to transform abstract knowledge into visual, interactive 3D models to assist students in learning and to build a dialogue and communication platform for interactive Q&A.

[0065] Specifically, during the implementation of the learning program, for abstract knowledge requiring immersive understanding, virtual reality (VR) technology is used to construct observable and operable learning methods. For example, in the study of celestial motion in physics, students wear VR devices to enter a virtual solar system, where they can adjust the Earth's orbital radius and rotation speed, and observe in real time the relationship between day and night, seasonal changes, tidal phenomena, and celestial motion. For knowledge requiring integration with real-world scenarios, augmented reality (AR) technology is used to overlay virtual learning. For example, in the study of solid geometry in mathematics, students use AR devices to scan cube illustrations in textbooks, instantly generating a 1:1 scale three-dimensional cube model in real space, which can be dragged and rotated with fingers to observe the correspondence between the front, side, and top views. During the implementation of the learning program, a dialogue and communication platform is established, supporting multi-form interaction and instant Q&A responses.

[0066] In this embodiment, virtual reality and augmented reality technologies are used to overcome the abstract barriers of traditional flat textbooks, helping students form intuitive spatial cognition and improve learning outcomes. At the same time, a dialogue and communication platform is built to provide students with interactive Q&A channels in natural language.

[0067] S400 retrieves the question text uploaded by students during the learning process, analyzes the learning intentions based on the question text, and extracts the core elements from the question text.

[0068] Specifically, the question text includes text input from the dialogue platform, speech-to-text results, and text attached to images (such as question descriptions). The dialogue platform performs standardized formatting on texts from different sources (e.g., removing special symbols and standardizing punctuation). Based on the processed question text, learning intentions are extracted, including types such as knowledge comprehension, application calculation, error correction and Q&A, and extended exploration. Core elements are extracted from the question text using named entity recognition and semantic role labeling technologies.

[0069] S500, based on student profile data, analyzed learning intentions, and extracted core elements, outputs answers that match students' cognitive levels through a large model.

[0070] Specifically, student profile data is the core basis for determining what students can understand and what depth of answers they need. It needs to focus on cognitive level-related characteristics. Learning intention is the core purpose of student questions, which needs to be mined from the question text using natural language processing technology to avoid merely relying on literal meanings and resulting in irrelevant answers. Core elements are the key entities and logical relationships extracted from the question text, which are the foundation for ensuring that the answers are comprehensive and accurate. Before generating answers, it is necessary to quantify students' cognitive levels based on the student profile data to avoid answer mismatch. The large model generates initial answers based on the synergistic constraints of cognitive level, learning intention, and core elements. These initial answers may have issues such as inaccuracy, insufficient fit, and awkward expression, requiring optimization to ensure answer quality.

[0071] In this embodiment, the core step of achieving accurate Q&A is to output answers that match the students' cognitive level through a large model based on student profile data, the analyzed learning intentions, and the extracted core elements. This breaks through the limitations of traditional question bank matching Q&A. Through multi-dimensional data fusion and intelligent generation by a large model, it ensures that the answers not only accurately solve the problem but also adapt to the students' abilities.

[0072] In one embodiment, such as Figure 2 As shown, S500 includes:

[0073] S520 assesses students’ knowledge coverage, accuracy of knowledge understanding, and ability to apply knowledge points based on student profile data, in order to quantify students’ knowledge level.

[0074] S540, based on knowledge level, learning intentions obtained from analysis, and extracted core elements, uses a large model to obtain initial answers that match the students' cognitive level;

[0075] S560 uses a pre-set domain knowledge base to perform retrieval-based verification on the initial answer to avoid errors, and uses a pre-set evaluation model to filter out awkward expressions in the initial answer before outputting the final answer.

[0076] Specifically, students' knowledge levels are quantified by assessing their knowledge coverage, accuracy of understanding knowledge points, and ability to apply knowledge points. In assessing knowledge coverage, the mastery rate of each knowledge point is first calculated based on test scores and homework data from student profiles. Then, the mastery rates of all knowledge points are combined to obtain the knowledge coverage. In assessing the accuracy of knowledge points, the accuracy rate of students' comprehension questions is statistically analyzed. Combined with past error types, the accuracy rate is categorized into three levels: high (accuracy rate ≥ 80%), medium (accuracy rate 60%-79%), and low (accuracy rate < 60%). In assessing the ability to apply knowledge points, the accuracy rate of students' answers to basic application questions, comprehensive application questions, and real-world scenario questions is statistically analyzed to calculate the application ability score. By combining these three dimensions—knowledge coverage, accuracy of understanding knowledge points, and ability to apply knowledge points—the knowledge level is divided into multiple levels, corresponding to different levels of answer depth. A rule base for the association of "learning intention - core elements - knowledge level" is established. Based on this rule base, an initial answer that matches the student's cognitive level is obtained through a large model. A domain knowledge base is established, and the core knowledge points in the initial answer are extracted. The corresponding index items are retrieved in the domain knowledge base, and the consistency between the answer content and the knowledge base information is compared. If inconsistencies are found, the correct information in the domain knowledge base is extracted, and the initial answer is corrected. At the same time, a preset evaluation model is used to determine whether the initial answer needs to be polished to filter out awkward expressions in the initial answer and output the final answer.

[0077] In one embodiment, developing personalized learning plans using a reinforcement learning model based on student profile data and learning objective data includes:

[0078] A knowledge point dependency graph is constructed based on teaching objective data. The knowledge point dependency graph includes the preconditions, parallel relationships, and progressive relationships of knowledge points. Based on the knowledge point dependency graph and student profile data, a personalized learning plan is formulated through a reinforcement learning model.

[0079] Specifically, a knowledge point dependency graph is a navigation map for determining the learning sequence. It needs to accurately reflect the prerequisite, parallel, and progressive relationships between knowledge points. Specifically, based on teaching objective data, the knowledge system of a subject is broken down into a three-level structure of "core knowledge point - sub-knowledge point - specific content." Based on this three-level structure, prerequisite relationships (i.e., knowledge point A must be mastered before learning knowledge point B), parallel relationships (i.e., knowledge point A and knowledge point B have no sequential relationship), and progressive relationships (i.e., knowledge point B is a deepening and expansion of knowledge point A) are defined. Based on knowledge... The knowledge point dependency graph is constructed by identifying the preceding, parallel, and progressive relationships between points and storing it in a graph database. Each knowledge point is a node, and the dependencies are edges. Node attributes include knowledge point ID, name, difficulty level, and suggested learning time, while edge attributes include relationship type and dependency strength (e.g., preceding relationship strength is 1.0, and progressive relationship strength is 0.8). Based on the knowledge point dependency graph and student profile data, personalized learning plans are developed using a reinforcement learning model. The learning plans must simultaneously satisfy both dependency relationships (logical order) and student characteristics (personalized needs).

[0080] In one embodiment, the process involves obtaining the text of questions uploaded by students during the learning process, parsing the learning intentions from the question text, and extracting the core elements from the question text, including:

[0081] The system standardizes and encodes the question text to generate an encoded feature representation of the current text; it obtains the encoded feature representation of historical dialogues, concatenates the encoded feature representation of the current text and the encoded feature representation of historical dialogues, captures the intent association through an attention mechanism to parse out the learning intent; and extracts the core elements from the question text through named entity recognition and semantic role labeling technology.

[0082] Specifically, the question text is standardized and encoded to generate an encoded feature representation of the current text. The student's learning intent needs to be determined in conjunction with the historical dialogue context. Specifically, past historical dialogue texts are extracted, and the same encoding method as the current text is used to generate vector features for each historical dialogue. Then, average pooling is applied to all historical dialogue vectors to obtain the encoded feature representation of the historical dialogue, which reflects the student's recent learning focus. The encoded feature representation of the current text is concatenated with the encoded feature representation of the historical dialogue to obtain a concatenated vector. A multi-head attention mechanism is then used to perform a linear transformation on the concatenated vector to obtain an intent-related feature vector. Finally, the intent-related feature vector is input into a fully connected layer to output the learning intent category. Named entity recognition and semantic role labeling technologies are used to extract core elements from the question text. Specifically, the BiLSTM-CRF model is used for named entity recognition. This model combines the sequence modeling capability of BiLSTM with the global optimal labeling capability of CRF, which can effectively identify professional entities in the field of education. The BERT-SRL model is used for semantic role labeling. This model is based on the semantic understanding capability of BERT and can accurately identify the "predicate-argument" relationship in the sentence (such as "calculate" is a predicate, "Pythagorean theorem" is an instrumental argument, and "staircase length" is a result argument).

[0083] In one embodiment, the intelligent learning companion method further includes:

[0084] For scarce disciplines, we employ small-sample learning and prompting engineering to activate the domain transfer capabilities of the large model through a small amount of question-and-answer data; we collect spoken language data from students in different regions to construct a dialect standard mapping table, and enhance the large model's tolerance for non-standard expressions through comparative learning.

[0085] Specifically, for scarce subjects, "knowledge point-question-answer" triplet data is collected. The collected data is filtered to obtain high-quality data, and each data point is labeled with the knowledge point category, question type, and answer difficulty level to ensure representativeness. Typically, the collected data only needs to meet the requirements of small-sample learning. Then, structured prompt templates are designed, integrating the small-sample data into the prompts to guide the large model in learning the knowledge and expression style of scarce subjects. Spoken language data from students in major dialect regions across the country is collected, covering both daily conversation and subject-specific question-and-answer language. Then, dialect experts and educators are organized to map the collected dialect data between dialect expressions and standard Mandarin expressions, obtaining a dialect standard mapping table. Based on the dialect standard mapping table, positive samples (standard expressions) and negative samples (non-standard expression variants) are constructed. Finally, the positive and negative samples are input into the large model, and the model is trained and its parameters optimized using a contrastive learning method. After training, the model can map semantically similar but differently expressed texts to a similar vector space, achieving accurate matching of synonymous texts with different forms.

[0086] In one embodiment, the intelligent learning companion method further includes:

[0087] By using a sequence pattern mining algorithm, the timestamp sequence of student behavior is analyzed to extract high-frequency learning behavior paths; based on the extracted high-frequency learning behavior paths, efficient and inefficient learning modes are distinguished; and students in inefficient learning modes are given real-time prompts.

[0088] Specifically, firstly, behavioral data of all students during the learning process is collected. This data includes students' login time, start and end times of learning knowledge points, time of completing assignments, time of asking questions, and time of participating in tests, etc., and the timestamp corresponding to each behavior is recorded to form a timestamp sequence of student behaviors. Then, a sequence pattern mining algorithm is used to analyze these timestamp sequences. First, the frequency of occurrence of individual behavior types is counted, and behaviors with a support (frequency of occurrence / total number of behaviors) ≥10% are selected to generate frequent itemsets. These frequent itemsets represent the more common individual learning behaviors among students. Next, based on the frequent itemsets, behavior sequences of length k (k≥2) are generated, and the support of each behavior sequence (number of occurrences / total number of behavior sequences) is calculated. Behavior sequences with a support ≥5% are selected as high-frequency learning behavior paths. Learning effectiveness indicators (such as knowledge point mastery rate, test scores, assignment accuracy, etc.) and time efficiency indicators (such as time spent completing the same learning task, number of knowledge points mastered per unit time, etc.) of students under different high-frequency learning behavior paths are collected. These indicators are quantitatively analyzed, and judgment criteria for high efficiency and low efficiency are set. Targeted prompts are generated in real time for students in low-efficiency learning mode. After the prompt is issued, the student's subsequent behavior is tracked (such as whether they click to jump to the resource, whether they start a new task, or whether they ask a question). If the student adjusts their behavior within 10 minutes (such as starting to watch the recommended video or doing new exercises), the prompt is deemed effective. If the student ignores the prompt three times in a row, or remains in an inefficient mode after adjustment, the prompt is upgraded (such as adding teacher intervention, with the subject teacher sending personalized guidance messages).

[0089] In one embodiment, after developing a personalized learning plan based on student profile data and teaching objective data using a reinforcement learning model, the method further includes:

[0090] Real-time collection of student learning feedback data, including accuracy rate of completing learning tasks, time spent, and satisfaction with proactive feedback; inputting the learning feedback data into the reinforcement learning model to update the learning plan.

[0091] Specifically, during the implementation of the learning program, a real-time data collection mechanism is established to continuously collect student learning feedback data, including: learning accuracy rate, which is the ratio of the number of correct answers to the total number of questions in a student's completed learning tasks (such as exercises, quizzes, etc.), reflecting the student's mastery of the knowledge points; time spent, which is the time spent by students to complete each learning task, reflecting the student's learning speed and efficiency; and satisfaction with proactive feedback, which is collected by setting up a satisfaction evaluation process during the learning process (such as having students rate and evaluate the learning content and learning methods after learning a knowledge point), to understand students' acceptance and preference for the learning program. The reinforcement learning model will periodically read the learning feedback data and adjust the model parameters and learning program accordingly.

[0092] In one embodiment, such as Figure 3 As shown, an intelligent learning companion device is provided, including: a data acquisition module 10, a learning plan formulation module 20, a knowledge learning module 30, a question text parsing module 40, and an answer output module 50, wherein:

[0093] The data acquisition module 10 is used to collect student profile data and define teaching objective data;

[0094] The learning plan development module 20 is used to develop personalized learning plans based on student profile data and teaching objective data through a reinforcement learning model. The learning plan includes the learning order of knowledge points and the corresponding learning time.

[0095] The knowledge learning module 30 is used to transform abstract knowledge into a visual and interactive 3D model through virtual reality and augmented reality technologies during the implementation of the learning program, in order to assist students in learning and build a dialogue and communication platform for interactive Q&A.

[0096] The question text parsing module 40 is used to obtain the question text uploaded by students during the learning process, parse out the learning intention based on the question text, and extract the core elements in the question text;

[0097] The answer output module 50 is used to output answers that match the students' cognitive level based on student profile data, the analyzed learning intentions, and the extracted core elements through a large model.

[0098] In one embodiment, the answer output module 50 is further configured to evaluate the student's knowledge learning coverage, accuracy of knowledge point understanding, and ability to apply knowledge points based on student profile data, so as to quantify the student's knowledge level; based on the knowledge level, the parsed learning intention, and the extracted core elements, an initial answer that matches the student's cognitive level is obtained through a large model; the initial answer is retrieval-based and verified through a preset domain knowledge base to avoid errors, and the awkward expressions in the initial answer are filtered through a preset evaluation model to output the final answer.

[0099] In one embodiment, the learning plan formulation module 20 is further configured to construct a knowledge point dependency graph based on teaching objective data, the knowledge point dependency graph including the preconditions, parallel relationships and progressive relationships of knowledge points; and formulate a personalized learning plan based on the knowledge point dependency graph and student profile data through a reinforcement learning model.

[0100] In one embodiment, the question text parsing module 40 is also used to standardize and encode the question text to generate the encoded feature representation of the current text; obtain the encoded feature representation of the historical dialogue, concatenate the encoded feature representation of the current text and the encoded feature representation of the historical dialogue, capture the intent association through the attention mechanism to parse out the learning intent; and extract the core elements from the question text through named entity recognition and semantic role labeling technology.

[0101] In one embodiment, the intelligent learning companion device also includes a model optimization module, which is used to activate the domain transfer capability of the large model by using small sample learning and prompting engineering for scarce subjects through a small amount of question and answer data; collect spoken language data of students from different regions, construct a dialect standard mapping table, and enhance the tolerance of the large model to non-standard expressions through comparative learning.

[0102] In one embodiment, the intelligent learning companion device further includes a learning mode differentiation module, which is used to analyze the timestamp sequence of student behavior through a sequence pattern mining algorithm to extract high-frequency learning behavior paths; differentiate between efficient and inefficient learning modes based on the extracted high-frequency learning behavior paths; and provide real-time prompts to students in inefficient learning modes.

[0103] In one embodiment, the intelligent learning companion device further includes a data feedback module for collecting students' learning feedback data in real time. The learning feedback data includes the accuracy rate of completing the learning, the time spent, and the satisfaction with proactive feedback. The learning feedback data is then input into a reinforcement learning model to update the learning plan.

[0104] Each module in the aforementioned intelligent learning companion device can be implemented entirely or partially through software, hardware, or a combination thereof.

[0105] In one embodiment, this application discloses a computer device including a memory and a processor. The memory stores a computer program that can run on the processor. When the processor loads the computer program, it executes an intelligent learning companion method as described in the above embodiment.

[0106] In one embodiment, this application discloses a computer-readable storage medium storing a computer program, wherein when the computer program is loaded by a processor, it executes an intelligent learning companion method as described above.

[0107] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. An intelligent learning method, characterized in that, The method comprises the following steps: Collecting student portrait data and clear teaching goal data; According to the student portrait data and the teaching goal data, a personalized learning plan is formulated through a reinforcement learning model, the learning plan including the learning order of knowledge points and the corresponding learning duration; During the execution of the learning plan, abstract knowledge is converted into visual and interactive three-dimensional models through virtual reality and augmented reality technology to assist students in learning and build a dialogue platform for interactive question answering; Obtaining the question text uploaded by the student during the learning process, analyzing the learning intention from the question text, and extracting the core elements in the question text; According to the student portrait data, the learning intention analyzed, and the core elements extracted, a large model is used to output an answer that meets the student's cognitive level; The method of formulating a personalized learning plan through a reinforcement learning model based on the student portrait data and the teaching goal data comprises the following steps: Constructing a knowledge point dependency graph based on the teaching goal data, the knowledge point dependency graph including the precedence relationship, parallel relationship and progressive relationship of knowledge points; According to the knowledge point dependency graph and the student portrait data, a personalized learning plan is formulated through a reinforcement learning model; The method of obtaining the question text uploaded by the student during the learning process, analyzing the learning intention from the question text, and extracting the core elements in the question text comprises the following steps: Standardizing and encoding the question text to generate the encoded feature representation of the current text; Obtaining the encoded feature representation of the historical dialogue, concatenating the encoded feature representation of the current text and the encoded feature representation of the historical dialogue, and capturing the intention association through an attention mechanism to analyze the learning intention; Extracting core elements from the question text through named entity recognition and semantic role labeling technology.

2. The intelligent learning method according to claim 1, wherein, The method of outputting an answer that meets the student's cognitive level through a large model based on the student portrait data, the learning intention analyzed, and the core elements extracted comprises the following steps: According to the student portrait data, the knowledge learning coverage, knowledge point understanding accuracy and knowledge point application ability of the student are evaluated to quantify the knowledge level of the student; According to the knowledge level, the learning intention analyzed, and the core elements extracted, an initial answer that meets the student's cognitive level is obtained through a large model; Through a pre-set domain knowledge base, the initial answer is checked for errors, and the harsh expressions in the initial answer are filtered through a pre-set evaluation model to output a final answer.

3. The intelligent learning method according to claim 1, wherein, The method further comprises the following steps: For scarce disciplines, small sample learning and prompt engineering are used to activate the domain migration ability of the large model through a small amount of question and answer data; Collecting spoken language corpora of students in different regions to construct a dialect standard mapping table, and enhancing the fault tolerance of the large model to non-standard expressions through contrastive learning.

4. The intelligent learning method according to claim 1, wherein, The method further comprises the following steps: Through a sequence pattern mining algorithm, the timestamp sequence of student behavior is analyzed, and a high-frequency learning behavior path is extracted; According to the extracted high-frequency learning behavior path, efficient and inefficient learning modes are distinguished; Real-time prompting is provided to students in the inefficient learning mode.

5. The intelligent learning method according to claim 1, wherein, The method further comprises: collecting learning feedback data of the student in real time, the learning feedback data including accuracy, time consumption, and satisfaction of active feedback; inputting the learning feedback data into the reinforcement learning model to update the learning scheme.

6. An intelligent learning companion device, comprising: The method comprises: a data acquisition module configured to collect student profile data and determine teaching target data; a learning scheme formulation module configured to formulate a personalized learning scheme through a reinforcement learning model based on the student profile data and the teaching target data, the learning scheme including a learning order of knowledge points and corresponding learning duration; a knowledge learning module configured to convert abstract knowledge into visual and interactive three-dimensional models through virtual reality and augmented reality technology to assist students in learning and build a dialogue platform for interactive question answering during execution of the learning scheme; a question text analysis module configured to obtain question text uploaded by the student during learning, analyze learning intention based on the question text, and extract core elements in the question text; an answer output module configured to output answers in accordance with the student profile data, the analyzed learning intention, and the extracted core elements through a large model to meet the cognitive level of the student; The learning scheme formulation module is further configured to construct a knowledge point dependency graph based on the teaching target data, the knowledge point dependency graph including a preposition relationship, a parallel relationship, and a progressive relationship between knowledge points; formulate a personalized learning scheme through a reinforcement learning model based on the knowledge point dependency graph and the student profile data; The question text analysis module is further configured to perform standardization processing and coding on the question text to generate coded feature representation of the current text; obtain coded feature representation of historical dialogues; splice the coded feature representation of the current text and the coded feature representation of the historical dialogues to capture intention association through an attention mechanism to analyze learning intention; and extract core elements from the question text through named entity recognition and semantic role labeling technology. 7.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-6 when the computer program is executed by the processor. The processor executes the computer program to implement the steps of the method of any one of claims 1 to 5.

8. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the method of any one of claims 1 to 5.

Images