Intelligent Agent heuristic question and answer teaching system based on knowledge graph driving

Abstract

The invention relates to the technical field of intelligent education, in particular to an intelligent Agent heuristic question and answer teaching system based on knowledge graph driving, and the system comprises a four-layer progressive architecture including a data layer, a knowledge layer, an Agent layer and an interaction layer: the data layer is responsible for the collection, preprocessing and privacy protection of multi-dimensional teaching data, and provides high-quality input for the construction of a knowledge graph; the knowledge layer is used for converting data into a structured knowledge graph and storing a multidisciplinary knowledge network in an entity-relationship-attribute triple form; in the scheme, a'data-knowledge-Agent-interaction 'four-layer architecture is constructed, knowledge structured association and dynamic updating are realized based on a knowledge graph, weak points of students are accurately positioned by means of sensing, reasoning and decision-making modules of an intelligent Agent, a personalized problem chain is generated, and a multi-modal interaction and privacy protection mechanism is combined, so that the knowledge knowledge association and dynamic updating are realized. Systematization of a knowledge system, personalization of question and answer guidance, precision of data utilization and dynamic knowledge updating are achieved, and the defects of a traditional teaching system are effectively overcome.

Description

Technical Field

[0001] This invention relates to the field of intelligent education technology, and in particular to a knowledge graph-driven intelligent agent-based heuristic question-answering teaching system. Background Technology

[0002] The intelligent agent-based heuristic question-and-answer teaching system is a teaching aid system that uses intelligent agents to simulate the interactive teaching process. Through heuristic methods such as questioning, guidance, and feedback, it helps students independently explore knowledge and solve problems. Its core is student-centered, stimulating learning initiative. Knowledge graph-driven systems use knowledge graphs as the underlying core support, constructing a structured network of subject knowledge in the form of "entity-relationship-attribute" triples. This makes the logical connections between knowledge points visible and reasonable, providing accurate knowledge basis for the intelligent agent's question-and-answer decisions.

[0003] Traditional teaching systems suffer from several shortcomings: First, knowledge is presented in a fragmented manner, often as isolated knowledge points, making it difficult to demonstrate the inherent logic between different pieces of knowledge (e.g., the dependency between factorization and solving quadratic equations in mathematics, or the causal relationship between the Opium War and the Self-Strengthening Movement in history is severed), hindering students from forming a systematic knowledge framework. Second, question-and-answer strategies are homogenized, failing to provide personalized guidance based on individual student knowledge gaps (e.g., some students do not understand discriminant calculations, while others cannot analyze text details), often relying on standardized answers and lacking heuristic, step-by-step guidance. Third, data utilization is inefficient, with superficial analysis of student behavioral data (e.g., types of incorrect answers, questioning history), making it difficult to accurately pinpoint weaknesses and raising concerns about data privacy. Fourth, knowledge updates are lagging, failing to quickly respond to textbook revisions, changes in curriculum standards, and other content updates, compromising the timeliness and accuracy of the knowledge system.

[0004] To address this, this invention proposes a knowledge graph-driven intelligent agent-based heuristic question-answering teaching system. By constructing a four-layer architecture of "data-knowledge-agent-interaction," it leverages the knowledge graph to achieve structured association and dynamic updates of knowledge. Utilizing the perception, reasoning, and decision-making modules of the intelligent agent, it accurately identifies students' weaknesses and generates personalized question chains. Combined with multimodal interaction and privacy protection mechanisms, it achieves a systematic knowledge system, personalized question-answering guidance, precise data utilization, and dynamic knowledge updates, effectively compensating for the shortcomings of traditional teaching systems. Summary of the Invention

[0005] Technical problems to be solved: Addressing the issues of fragmented knowledge, homogenized questions and answers, inefficient data utilization, and insufficient privacy protection in traditional teaching systems.

[0006] To address the shortcomings of existing technologies, this invention provides a knowledge graph-driven intelligent agent-based heuristic question-answering teaching system, thereby solving the technical problems mentioned in the background section.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0008] A knowledge graph-driven intelligent agent-based heuristic question-answering teaching system, comprising a four-layer progressive architecture: data layer, knowledge layer, agent layer, and interaction layer.

[0009] The data layer is responsible for the collection, preprocessing, and privacy protection of multi-dimensional teaching data, providing high-quality input for knowledge graph construction;

[0010] The knowledge layer transforms data into a structured knowledge graph, storing a multidisciplinary knowledge network in the form of "entity-relationship-attribute" triples;

[0011] The Agent layer relies on knowledge graphs to generate reasoning and question-answering strategies, and achieves personalized heuristic question answering through the collaborative efforts of perception, reasoning, decision-making, and learning modules.

[0012] The interaction layer serves as the user interface, connecting the Agent and the student to complete multimodal heuristic question-and-answer interaction.

[0013] In one possible implementation, the core data types of the data layer include subject resource data, knowledge-related data, and user behavior data; the data collection methods cover obtaining structured data through API integration, obtaining unstructured data using web crawlers and OCR technology, and recording user behavior data through interactive layer tracking; the data preprocessing process includes cleaning, standardization, and association integration, and data privacy is protected through anonymization, a federated learning framework, and other means.

[0014] In one possible implementation, the core elements of the knowledge graph in the knowledge layer include entities, relationships, and attributes. Entities cover science entities, humanities entities, and general entities. Relationships include dependency relationships, inclusion relationships, causal relationships, and interdisciplinary associations. Attributes are feature descriptions of entities. The knowledge graph construction steps include entity extraction using a BERT model combined with a subject dictionary, relationship extraction using a rule engine and a machine learning model, and visualization storage using the Neo4j graph database. The knowledge graph also has a dynamic update mechanism that includes automatic updates and manual corrections.

[0015] In one possible implementation, the perception module of the Agent layer is used to parse student input and transform it into queries that can be understood by the knowledge graph; the reasoning module infers the student's knowledge gaps based on the knowledge graph and user data; the decision-making module generates a tiered heuristic question chain to guide the student from the known to the unknown; and the learning module optimizes the question-answering strategy through reinforcement learning, using the improvement of student answer accuracy, question chain completion time, and knowledge point mastery rate as positive reward indicators, and the continuous wrong answers to the same type of questions and feedback of "question irrelevant" as negative indicators.

[0016] In one possible implementation, the core interface of the interaction layer includes a question-and-answer interface, a knowledge graph visualization interface, a learning report interface, and a personalization settings interface; the question-and-answer interface supports multimodal input and displays agent questions and knowledge graph fragments; the knowledge graph visualization interface supports node zooming and clicking to view details; the learning report interface displays knowledge mastery and ability changes in charts; and the personalization settings interface allows students to select interaction modes, question difficulty preferences, and knowledge graph display methods.

[0017] In one possible implementation, the association and integration step in the data preprocessing of the data layer specifically involves establishing a "test question-knowledge point-ability dimension" mapping table to associate test questions with corresponding knowledge points and ability dimensions.

[0018] In one possible implementation, the relationships in the knowledge layer specifically include: the dependency relationship in the field of mathematics that "factorization is the basis for solving quadratic equations in one variable", the inclusion relationship in the field of language that "rhetorical devices include metaphor and personification", the causal relationship in the field of history that "the defeat in the Opium War led to the rise of the Self-Strengthening Movement", and the interdisciplinary connection that "mathematical function graphs can be used to analyze the trajectory of uniformly accelerated motion in physics".

[0019] In one possible implementation, the decision module of the Agent layer generates the heuristic question chain in a step-by-step manner, that is, starting from the student's known knowledge and gradually moving towards unknown knowledge.

[0020] Beneficial effects compared to existing technologies:

[0021] This solution constructs a four-layer architecture of "data-knowledge-agent-interaction" and achieves close linkage of data flows at each layer. This enables the system to accurately integrate teaching resources and deeply integrate personalized teaching guidance. The data layer collects subject resources, knowledge associations, and user behavior data from multiple dimensions, and provides high-quality data support for teaching through standardized processing and privacy protection mechanisms. This solves the problems of data dispersion, low utilization rate, and privacy security in traditional teaching. The knowledge layer constructs an interdisciplinary knowledge graph using "entity-relationship-attribute" triples, combined with a dynamic update mechanism, to ensure the accuracy and timeliness of the knowledge system, making up for the shortcomings of vague knowledge associations and lagging updates in traditional teaching.

[0022] The Agent layer, through the collaboration of perception, reasoning, decision-making, and learning modules, accurately identifies students' knowledge gaps and generates a tiered chain of questions. For example, it designs guiding questions that lead from formula review to logical connections for students with weak points in mathematical discriminant analysis, realizing a shift from "cramming" teaching to "heuristic" guidance. The multimodal interface of the interaction layer supports input methods such as text, voice, and handwriting. Combined with knowledge graph visualization and personalized settings, it adapts to the needs of students with different learning styles, improving the convenience and effectiveness of learning interaction.

[0023] In summary, by deeply integrating knowledge graphs and intelligent agents, the system achieves both structured and dynamic management of the knowledge system and personalized, precise, and heuristic teaching, significantly improving students' learning efficiency and depth of knowledge mastery. Attached Figure Description

[0024] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, the preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0025] Figure 1 This is a schematic diagram of the system framework of the present invention. Detailed Implementation

[0026] Preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. However, the present invention can also be implemented in various different forms, and therefore the present invention is not limited to the embodiments described below.

[0027] The technical solution in this application embodiment is to solve the problems mentioned in the background art, and the overall idea is as follows:

[0028] Example:

[0029] Please refer to Figure 1As shown, this embodiment introduces a knowledge graph-driven intelligent agent-based heuristic question-answering teaching system. The system adopts a four-layer progressive architecture of "data-knowledge-agent-interaction," with each layer closely linked through data flow to form a closed-loop teaching system. Specifically, the data layer is the "raw material library," providing the basic data for constructing the knowledge graph; the knowledge layer is the "processing workshop," transforming the data into a structured knowledge graph; the agent layer is the "decision-making center," relying on the knowledge graph to realize reasoning and question-answering strategy generation; and the interaction layer is the "user interface," connecting the agent and the student to complete heuristic question-answering interaction. Details are as follows:

[0030] (I) Data Layer: Collection, Preprocessing, and Privacy Protection of Teaching Data

[0031] The data layer is responsible for collecting multi-dimensional teaching data and providing high-quality input for knowledge graph construction through standardized processing, while strictly protecting user privacy.

[0032] 1 Core Data Types

[0033] Subject-specific resource data: textbook texts, test question banks, teaching video scripts, etc.; for example: textbook definition, example analysis, and after-class exercises for "quadratic equations in one variable" in junior high school mathematics; explanation of the elements and model excerpts for "narrative writing" in Chinese language; background and impact analysis of the "Xinhai Revolution" in history.

[0034] Knowledge-related data: The logical relationships between knowledge points, labeled by subject matter experts or extracted using NLP technology; for example: in mathematics, "factorization is the foundation for solving quadratic equations in one variable"; in Chinese, "the role of metaphor is to enhance the vividness of language"; in history, "the Self-Strengthening Movement laid the foundation for modern industry".

[0035] User behavior data: Students' answer records (e.g., "wrong answer was x") 2 Factorize 5x + 6 = 0, ask questions (e.g., "Why is the detail of the father buying oranges in 'The Back View' important?"), study time, interactive feedback (e.g., "I think a certain question is too difficult"), etc.

[0036] 2 Data Acquisition Methods

[0037] Structured data: Access electronic versions of textbooks and standardized test question banks by connecting to educational publishers' databases via API;

[0038] Unstructured data: Web crawlers were used to collect publicly available teaching resources (such as subtitles for Khan Academy teaching videos and courseware from the National Smart Education Platform for Primary and Secondary Schools), and OCR technology was used to process student handwritten answer images and classroom note photos into parsable text;

[0039] User behavior data: Automatically recorded through event tracking in the interaction layer, such as the time students spend on the answer interface, the number of times they modify their answers, and their likes / favorites on questions;

[0040] 3 Data Preprocessing Flow

[0041] Cleaning: Remove duplicate data (such as the same question appearing repeatedly in different test papers) and correct incorrect labels (such as correcting the incorrect association of "Pythagorean theorem" with "triangle area formula" to "right triangle side length calculation").

[0042] Standardization: Unify data formats, such as converting all date data to the "year-month-day" format, standardizing the difficulty of test questions into three levels of "easy / medium / difficult", and quantifying the "answer score" of Chinese reading comprehension into "0-10 points";

[0043] Linkage and Integration: Establish a mapping table of "test questions - knowledge points - ability dimensions", for example, mapping test questions "solve x" to "knowledge points - ability dimensions". 2 The phrase "-3x+2=0" is related to the knowledge point of "factorization of quadratic equations" and the dimension of "computation ability"; the phrase "analyzing the detailed descriptions in 'The Back View'" is related to the knowledge point of "the function of details in narrative texts" and the dimension of "textual analysis ability".

[0044] 4. Data Privacy Protection

[0045] User behavior data is anonymized by removing personally identifiable information (such as name, student ID, and school) and retaining only learning behavior characteristics (such as types of incorrect answers and tags for mastering knowledge points).

[0046] Using a federated learning framework, student terminals only upload encrypted feature data (such as "factorization error rate 0.3"), while the original data is stored locally to avoid leakage of sensitive information.

[0047] Strictly adhering to relevant laws and regulations, users can independently select the scope of data usage (such as "for personalized recommendations only" or "not for model training") through the "Privacy Settings" in the interactive layer, and one-click export or deletion of personal data is supported;

[0048] (II) Knowledge Layer: Construction and Dynamic Updating of Interdisciplinary Knowledge Graphs

[0049] Knowledge graphs are the "knowledge brain" of a system, storing multidisciplinary knowledge networks in the form of "entity-relationship-attribute" triples, and supporting general use in arts and sciences.

[0050] 1. Core Elements of Knowledge Graph

[0051] entity:

[0052] Entities in the sciences: such as the "quadratic equation in one variable" and "discriminant" in mathematics, and the "Newton's second law" and "acceleration" in physics;

[0053] Humanities subjects: such as "rhetorical devices" and "classical Chinese function words" in Chinese language, and "the Opium War" and "the Self-Strengthening Movement" in history;

[0054] General entities: such as "example problem", "solution steps", "ability objectives (such as calculation ability, text analysis ability)";

[0055] Relationship: The logical connection between entities, including:

[0056] Dependency relationships (such as "the quadratic formula depends on the discriminant" and "the emotional understanding of 'The Back View' depends on the analysis of detailed descriptions");

[0057] Inclusion relationship (e.g., "a quadratic equation in one variable includes the general form and the vertex form"; "rhetorical devices include metaphor and personification");

[0058] Causal relationships (e.g., "Δ>0 leads to the equation having two unequal real roots" or "The defeat in the Opium War led to the rise of the Self-Strengthening Movement");

[0059] Interdisciplinary connections (e.g., "the graph of a mathematical function can be used to analyze the trajectory of uniformly accelerated motion in physics" and "the timeline of historical events can be compared with the narrative order of a narrative text in Chinese");

[0060] Attributes: Characteristic descriptions of an entity, such as:

[0061] The attributes of "Quadratic Equations in One Variable" are: Difficulty "Medium", Applicable Grade Level "Junior High School", Related Example Problems "3", Corresponding Ability "Calculation Ability";

[0062] The "Detailed Description of 'The Back View'" is categorized as follows: Difficulty: Medium; Applicable Grade Level: High School; Related Exam Point: Emotional Expression; Corresponding Ability: Text Analysis Skills.

[0063] 2. Knowledge Graph Construction Steps

[0064] Entity extraction: Using the BERT model combined with a subject-specific dictionary, knowledge points are identified from the textbook text; for example:

[0065] From mathematical texts, "of the form ax" 2 The equation +bx+c=0 (a≠0) is called a quadratic equation in one variable. Extract the entity "quadratic equation in one variable" from the following:

[0066] Extract the entity "metaphor (rhetorical device)" from the Chinese text "metaphor is a device that uses something that is similar to thing A to describe or explain thing A";

[0067] Relationship extraction: Identifying relationships through a rule engine and machine learning model; for example:

[0068] Extract the "solution dependency" relationship between "quadratic equation" and "factorization" from the mathematical text "When solving a quadratic equation, it can be first transformed into two linear equations through factorization".

[0069] Extract the causal relationship between the Self-Strengthening Movement and the Opium War from the historical text that "the Self-Strengthening Movement advocated learning Western technology, and its rise was closely related to the deepening national crisis after the Opium War";

[0070] Visualized storage: It uses Neo4j graph database for storage. Triples (entity 1, relation, entity 2) are stored in the form of nodes and edges, and it supports visual queries (such as the relation graph of "quadratic equation → solution dependency → factorization" can be displayed intuitively).

[0071] 3. Knowledge Graph Dynamic Update Mechanism

[0072] Automatic updates: Monitor changes in subject resources (such as textbook revisions and the release of new curriculum standards), identify new knowledge points through text comparison (such as the new example of "cross multiplication method" in the new mathematics textbook), automatically add them to the graph and establish connections; track frequently asked questions by students (such as "the specific impact of the ancient Silk Road on the exchange between Eastern and Western civilizations"), and recommend them to subject experts for review and supplementation as new entities;

[0073] Manual correction: Subject teachers can correct errors through the back-end review interface. For example, the incorrect relationship between "discriminant and number of roots" "Δ=0→no real roots" can be corrected to "Δ=0→two equal real roots"; history teachers can add the "direct causal relationship" between "the Second Opium War and the destruction of the Old Summer Palace".

[0074] (III) Agent Layer: Optimization of Reasoning and Question Answering Strategies of Intelligent Agents

[0075] The intelligent agent is the system's "teaching decision-maker," working collaboratively through four modules: perception, reasoning, decision-making, and learning to achieve personalized, heuristic question-and-answer sessions.

[0076] 1. Core Module Functions

[0077] Perception module: Parses student input and transforms it into queries that can be understood by the knowledge graph;

[0078] Example: A student asks, "Why x..." 2 "+1=0 has no real roots?" The perception module, through word segmentation and intent recognition, transforms this into a standardized query: "Query 'quadratic equation x'". 2 The relationship between +1=0' and 'existence of real roots' and the reasoning basis;

[0079] For example, when a student asks, "What is the purpose of the detail of the father buying oranges in 'The Back View'?", the perception module transforms it into "Querying the relationship between 'The Back View + Father Buying Oranges + Detailed Description' and 'Emotional Expression'".

[0080] Reasoning module: Infers students' knowledge gaps based on knowledge graphs and user data;

[0081] Example: Processing "x" 2 When encountering the problem "+1=0 has no real roots", the reasoning module calls the knowledge graph for "Δ=b". 2 Based on the relationship "-4ac<0→no real roots", and combined with the historical data of "not mastering the discriminant calculation" in the students' answer records, the weak point was identified as "the logical connection between the discriminant and the existence of real roots".

[0082] For example, when dealing with the detailed analysis of "The Back View", the reasoning module calls the relationship of "detailed description → reflecting the character's emotions", and combined with the student's historical error "once missed the dimension of 'father-son affection'", the weak point is located "the analytical angle of the connection between details and emotions".

[0083] Decision Module: Generates a tiered, heuristic chain of questions to guide students from the known to the unknown; for example, to address weaknesses in "discriminative relational logic," a question chain is generated:

[0084] What is the discriminant formula for a quadratic equation? (Review basic formulas)

[0085] “x 2 In the equation +1 = 0, what are the values ​​of a, b, and c respectively? (Guided by substitution calculation)

[0086] "Calculate the value of Δ for this equation?" (Enhancing calculation steps)

[0087] "What are the characteristics of the roots of the equation when Δ < 0?" (Related to core logic)

[0088] Example: Targeting the weakness in "the connection between details and emotions," generate a chain of questions:

[0089] "When Father buys oranges, he needs to cross the platform. Is this a difficult action?" (Guide the child to observe the details themselves.)

[0090] "The father is old and in poor health, yet he insists on doing this action. What attitude does this reflect about his son?" (Related to emotions)

[0091] "Does this detail make you feel that the author's feelings towards his father have changed?" (Extending the main theme of the text)

[0092] Learning module: Optimizes question-answering strategies through reinforcement learning to improve adaptability;

[0093] Reward Mechanism: Positive indicators include "student's correct answer rate" (e.g., +5 points for a correct answer rate ≥ 80%), "time to complete the question chain" (e.g., +3 points for a time ≤ 5 minutes), and "improvement in knowledge point mastery rate" (e.g., +10 points for an improvement from 30% to 60%); negative indicators include "students answering the same type of question incorrectly more than 3 times" (-2 points per instance) and "feedback that the question is irrelevant" (-5 points).

[0094] Strategy Adjustment: If the reward score for a certain questioning method (such as directly asking "discriminant formula") remains below the threshold (e.g., <3 points), automatically switch to contextualized questioning (e.g., "In equation x..."). 2 In the equation -4x+4=0, how do you determine the root using the discriminant? If students respond poorly to abstract questions about the "function of detailed descriptions," change the approach to questions that relate to specific texts (e.g., "Does the description of the father's 'staggering' movements remind you of his physical condition?").

[0095] 2. Examples of Reasoning Logic

[0096] Science Case Study: A student asked, "How can I solve x using factorization?" 2 +5x+6=0”;

[0097] The reasoning module locates the relationship between "factorization method" and "solution method" of "quadratic equation in one variable" in the knowledge graph, and extracts the step chain "constant term decomposition → matching coefficient of linear term → conversion to product form";

[0098] By linking the student's historical data, it was discovered that he had previously misdivided "x" 2 +3x+2” (2 is decomposed into 1×2, but 1+2=3 is not verified). The weakness is judged to be “the matching logic between the decomposition of the constant term and the coefficient of the linear term”.

[0099] The decision module generates a chain of questions:

[0100] “x 2 What are the coefficients of the constant term and the linear term in +5x+6? (Clarify the core elements)

[0101] "What two sets of integers can the constant term 6 be factored into?" (List the possibilities)

[0102] Which set of numbers sums to a linear term with a coefficient of 5? (Strengthening the matching logic)

[0103] "Please write down the factorized form and solve?" (Application verification)

[0104] Humanities Case Study: A student asks, "Why did the Self-Strengthening Movement fail to save the Qing Dynasty?"

[0105] The reasoning module calls the relationship chain in the knowledge graph: "Westernization Movement → Reasons for failure → Constraints of the feudal system, dependence on the West for technology, and failure to reform the system";

[0106] Based on the student's answer to the history question that "they only answered 'technologically backward'", the weakness was identified as "the ability to analyze the reasons for the failure of historical events from multiple perspectives".

[0107] The decision module generates a chain of questions:

[0108] What were the main measures of the Self-Strengthening Movement? (e.g., military, industry, education) (Reviewing basic historical facts)

[0109] Did these measures change the feudal system of the Qing Dynasty? (Analysis from the perspective of guiding institutions)

[0110] "Where did the technology and equipment of the Westernization enterprises mainly rely? What problems would this lead to?" (Guiding the perspective of external dependence)

[0111] "Considering all these factors, what do you think is the root cause?" (Summary and Improvement)

[0112] (iv) Interaction Layer: Multimodal Heuristic Question-Answering Interaction

[0113] The interaction layer serves as a "dialogue window" between students and the agent, enabling natural interaction through a multimodal interface and supporting the needs of students with different learning styles.

[0114] 1. Core Interface Design

[0115] Question-and-answer interface: The left side displays the Agent's question (e.g., "What two numbers can 6 be factored into?"), the right side supports multimodal input (text, voice, handwriting), and the bottom displays knowledge graph fragments (e.g., a node relationship diagram of "factorization steps"); for example, a student can handwrite "x 2 The system converts the "+5x+6" decomposition process into text through image recognition and analysis.

[0116] Knowledge graph visualization interface: Students can zoom in and out and click on nodes to view details (e.g., clicking the "Discriminant" node displays formulas, examples, and common mistakes); in the history subject, clicking the "Opium War" node displays a timeline diagram showing the relationship between "cause → process → result → impact";

[0117] Learning report interface: The radar chart shows the knowledge mastery level (e.g., the mastery rate of knowledge points related to "quadratic equation in one variable" in mathematics: factorization 70%, quadratic formula 50%), and the line graph shows the change in ability (e.g., "text analysis ability" increased from 30 points to 60 points). Weak points and improvement suggestions are marked (e.g. "Suggestion to strengthen discriminant calculation practice").

[0118] Personalized settings interface: Supports students to select interaction mode ("voice Q&A", "text Q&A"), question difficulty preference ("basic" "challenging"), and knowledge graph display method ("simple version" "detailed version");

[0119] 2. Example of Interaction Flow

[0120] Science Case Study (Factorization):

[0121] Students entered "How to solve x using factorization" on the question-and-answer interface. 2 +5x+6=0 (Supports voice input "Solve x squared plus 5x plus 6 equals 0");

[0122] The interaction layer passes the question to the Agent perception module, and the Agent generates the first question "x". 2 What is the constant term in +5x+6? (Please show the answer.)

[0123] The student writes "6" by hand and submits it. The interaction layer confirms its correctness through image recognition. The Agent then generates the second question: "What two integers can be factored into the product of 6?"

[0124] The student answers "2 and 3", and the agent responds, "Correct. Is the sum of these two numbers equal to the coefficient of the linear term, which is 5?"

[0125] The student answers "yes", and the agent guides the student to write out the decomposition form. Finally, the student derives the solution as x = -2 or x = -3.

[0126] The interactive layer displays the complete knowledge path of "solving quadratic equations in one variable by factorization" and pushes three similar practice questions (with different difficulties);

[0127] Humanities Case Study (Detailed Analysis of "The Back View"):

[0128] A student asked, "What is the purpose of the detail in 'The Back View' where the father buys oranges?"

[0129] The agent generates the first question: "When the father buys oranges, he needs to cross the platform. Is this an easy task for him?"

[0130] The student answered in voice, "It's not easy, because he's in poor health," and the system transcribed the message into text and confirmed it.

[0131] The agent continued by asking, "What emotion does this reflect in the father's feelings for his son?" The student answered, "Love."

[0132] The agent responded, "Very good. Now think about how this detail changes the author's feelings towards his father?" guiding students to analyze the text in conjunction with its main theme.

[0133] The interactive layer displays a knowledge graph of "detailed description → emotional expression" and pushes a comparative analysis exercise on the details of the father building the house in "The Steps".

[0134] In summary, taking "students learning to solve quadratic equations using factorization" as an example, the complete system operation process is as follows:

[0135] 1. Data layer preparation: Collect the definition and examples of "factorization" from the textbook, and record students' historical errors (such as "incorrectly factoring x"). 2 +3x+2”), and then anonymize and standardize it;

[0136] 2. Knowledge layer construction: Extract the entities "factorization method" and "quadratic equation in one variable", establish "solution dependency" relationship, and label the attribute "difficulty: medium";

[0137] 3. Agent-level reasoning and decision-making:

[0138] The perception module interprets the student's question as "find the solution to x using factorization". 2 The steps to get +5x+6=0;

[0139] The reasoning module uses historical data to pinpoint the weakness in "constant term decomposition and linear term coefficient matching";

[0140] The decision-making module generates a hierarchical chain of questions.

[0141] 4. Interaction Layer: Students answer questions via text / voice, the Agent provides real-time feedback, and ultimately guides students to deduce the answers independently;

[0142] 5. Effectiveness Evaluation:

[0143] Short-term assessment: Compare students' accuracy rate (from 30% to 80%) and time to complete the question chain (from 10 minutes to 5 minutes);

[0144] Long-term assessment: After three practice sessions, the mastery rate of this knowledge point remained stable at 90%, and the accuracy rate of similar questions in the unit test improved (from 50% to 85%).

[0145] Subjective assessment: Student feedback was collected through questionnaires regarding "whether the question chain helps with understanding" (80% of students chose "helpful"), and teacher feedback regarding "whether it reduces the tutoring burden".

[0146] Finally, it should be noted that the above embodiments are merely examples for clearly illustrating the present invention and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A knowledge graph-driven intelligent agent-based heuristic question-answering teaching system, characterized in that: The system comprises a four-layer progressive architecture: data layer, knowledge layer, agent layer, and interaction layer. The data layer is responsible for the collection, preprocessing, and privacy protection of multi-dimensional teaching data, providing high-quality input for knowledge graph construction; The knowledge layer transforms data into a structured knowledge graph, storing a multidisciplinary knowledge network in the form of "entity-relationship-attribute" triples; The Agent layer relies on knowledge graphs to generate reasoning and question-answering strategies, and achieves personalized heuristic question answering through the collaborative efforts of perception, reasoning, decision-making, and learning modules. The interaction layer serves as the user interface, connecting the Agent and the student to complete multimodal heuristic question-and-answer interaction.

2. The knowledge graph-driven intelligent agent-based heuristic question-answering teaching system as described in claim 1, characterized in that, The core data types of the data layer include subject resource data, knowledge association data, and user behavior data; Data collection methods include obtaining structured data through API integration, obtaining unstructured data using web crawlers and OCR technology, and recording user behavior data through interactive layer tracking. The data preprocessing process includes cleaning, standardization, and correlation integration, and uses anonymization, a federated learning framework, and data privacy protection.

3. The knowledge graph-driven intelligent agent-based heuristic question-answering teaching system as described in claim 1, characterized in that, The core elements of the knowledge graph in the knowledge layer include entities, relations, and attributes. Entities cover science entities, humanities entities, and general entities. Relationships include dependency relations, inclusion relations, causal relations, and interdisciplinary associations. Attributes are feature descriptions of entities. The knowledge graph construction steps include entity extraction using the BERT model combined with a subject dictionary, relation extraction using a rule engine and machine learning model, and visualization storage using the Neo4j graph database; and the knowledge graph has a dynamic update mechanism that includes automatic updates and manual corrections.

4. The knowledge graph-driven intelligent agent-based heuristic question-answering teaching system as described in claim 1, characterized in that, The perception module of the Agent layer is used to parse student input and transform it into queries that can be understood by the knowledge graph; the reasoning module infers the student's knowledge gaps based on the knowledge graph and user data. The decision-making module generates a tiered, heuristic chain of questions to guide students from the known to the unknown. The learning module uses reinforcement learning to optimize question-answering strategies by using positive reward indicators such as the accuracy of students' answers, the time taken to complete the question chain, and the mastery of knowledge points, and negative indicators such as consecutive incorrect answers to the same type of questions and feedback that the question is irrelevant.

5. The knowledge graph-driven intelligent agent-based heuristic question-answering teaching system as described in claim 1, characterized in that, The core interfaces of the interaction layer include a question-and-answer interface, a knowledge graph visualization interface, a learning report interface, and a personalization settings interface. The question-and-answer interface supports multimodal input, displaying agent-asked questions and knowledge graph fragments; The knowledge graph visualization interface supports node zooming and clicking to view details; The learning report interface uses charts to display changes in knowledge mastery and abilities; the personalized settings interface allows students to select interaction modes, question difficulty preferences, and knowledge graph display methods.

6. The knowledge graph-driven intelligent agent-based heuristic question-answering teaching system as described in claim 2, characterized in that, The association and integration step in the data preprocessing of the data layer specifically involves establishing a "test question-knowledge point-ability dimension" mapping table to associate test questions with corresponding knowledge points and ability dimensions.

7. The knowledge graph-driven intelligent agent-based heuristic question-answering teaching system as described in claim 3, characterized in that, The relationships within the knowledge layer specifically include: the dependency relationship in the field of mathematics that "factorization is the foundation for solving quadratic equations in one variable"; the inclusion relationship in the field of language that "rhetorical devices include metaphor and personification"; the causal relationship in the field of history that "the defeat in the Opium War led to the rise of the Self-Strengthening Movement"; and the interdisciplinary connection that "mathematical function graphs can be used to analyze the trajectory of uniformly accelerated motion in physics".

8. The knowledge graph-driven intelligent agent-based heuristic question-answering teaching system as described in claim 4, characterized in that, The decision-making module of the Agent layer generates a heuristic question chain in a step-by-step manner, that is, starting from the student's known knowledge and gradually advancing towards unknown knowledge.

Images