Dynamic text response method and system for educational question-answering robot

Abstract

The present invention provides a dynamic text response method and system for an educational question-and-answer robot, which relates to the field of data processing technology. The method includes: after the educational question-and-answer robot matches the target guidance skeleton according to the student's real-time question, it locates the starting guidance node; fills the static guidance language with question-and-answer scenario features and outputs the starting dynamic guidance language; iteratively performs cross-level dynamic matching and binding guidance language filling until the real-time educational goal is closed; locates real-time weak nodes based on the learning situation portrait and text response tracking data; loads it into the educational goal association topology, and narrows and locates the educational association topology nodes; and performs random guidance on students until the associated educational goal is closed. The present invention solves the technical problem that in traditional educational question-and-answer robots, the guidance language is usually generated based on a fixed template, and the guidance content cannot be dynamically adjusted according to the student's specific learning situation or answer feedback, resulting in unsatisfactory learning results.

Description

Technical Field

[0001] The present invention relates to the field of data processing, and more specifically, to a dynamic text response method and system for an educational question-and-answer robot. Background Art

[0002] With the in-depth application of artificial intelligence in education, educational question-and-answer robots have gradually become an important tool for personalized learning, designed to accurately respond to students' questions and provide learning guidance through intelligent interaction. However, existing technologies, such as the question-and-answer system described in patent CN110209773A, suffer from fundamental educational adaptation flaws: their mechanical node jump mechanism relies on preset service interface call logic, unable to dynamically guide knowledge mastery based on students' real-time cognitive state, resulting in the ineffectiveness of guidance for complex questions. Furthermore, the system processes interactive nodes in isolation, neither integrating historical data from error books with real-time behavioral characteristics nor lacking cross-disciplinary correlation analysis capabilities, making it difficult to accurately locate systemic weaknesses in learning.

[0003] More importantly, the traditional system generates guidance using fixed templates, and the learning path rigidly follows the order of the textbooks. It is unable to dynamically adjust the content or path based on student feedback, which leads to frustration of learning interest and autonomy. Its termination condition is only the return of the service transaction result. It is separated from the closed loop of educational goal verification and its essence is still a crude application of the industrial process framework to the educational scenario. It is in urgent need of breakthroughs through dynamic skeleton calibration driven by learning conditions and interdisciplinary weak point positioning. Summary of the Invention

[0004] In view of the above problems, the purpose of the present invention is to provide a dynamic text response method and system for an educational question-answering robot.

[0005] A first aspect of the present invention provides a dynamic text response method for an educational question-and-answer robot, the method comprising: after the educational question-and-answer robot matches a target guidance skeleton corresponding to a real-time educational goal according to a real-time question asked by a student, the robot locates a starting guidance node in the target guidance skeleton; based on the real-time question asked by the student, the static guidance words bound to the starting guidance node are filled with question-and-answer scene features, and a starting dynamic guidance word is output; based on the student's response text to the starting dynamic guidance word, the educational question-and-answer robot iteratively performs cross-level dynamic matching of guidance nodes and binding guidance word filling in the target guidance skeleton until a closed loop is achieved for the real-time educational goal; calling text response tracking data, and locating real-time weak nodes based on the learning situation portrait and the text response tracking data analysis; loading the real-time weak nodes and the real-time educational goal into the educational goal association topology, and narrowing and locating the educational association topology node; extracting the associated guidance skeleton of the educational association topology node to perform random guidance on the student until a closed loop is achieved for the associated educational goal.

[0006] A second aspect of the present invention provides a dynamic text response system for an educational question-and-answer robot, which is used for the dynamic text response method of the educational question-and-answer robot mentioned above. The system includes: a starting guide node positioning module, which is used for the educational question-and-answer robot to locate the starting guide node in the target guide skeleton after matching the target guide skeleton corresponding to the real-time educational goal according to the student's real-time question; a question-and-answer scene feature filling module, which is used for filling the static guide language bound to the starting guide node with question-and-answer scene features according to the student's real-time question, and outputting the starting dynamic guide language; a binding guide language filling module, which is used for the educational question-and-answer robot to iteratively perform cross-level dynamic matching of guide nodes and binding guide language filling in the target guide skeleton based on the student's response text to the starting dynamic guide language until the real-time educational goal is closed; a real-time weak node positioning module, which is used to call text response tracking data and locate the real-time weak node based on the learning situation portrait and the text response tracking data analysis; an associated topology node positioning module, which is used to load the real-time weak node and the real-time educational goal into the educational goal associated topology, and narrowly locate the educational associated topology node; and a random guidance module, which is used to extract the associated guidance skeleton of the educational associated topology node and perform random guidance on the student until the associated educational goal is closed.

[0007] A third aspect of the present invention provides a computer-readable storage medium.

[0008] In order to solve the above technical problems, the technical solutions of the present invention are as follows:

[0009] After matching the target guidance skeleton corresponding to the real-time education goal according to the real-time question asked by the student, the educational question-answering robot locates the starting guidance node in the target guidance skeleton; fills the static guidance words bound to the starting guidance node with question-answering scene features according to the real-time question asked by the student, and outputs the starting dynamic guidance words; based on the student's response text to the starting dynamic guidance words, the educational question-answering robot iteratively performs cross-level dynamic matching of guidance nodes and binding guidance word filling in the target guidance skeleton until the real-time education goal is closed; calls text response tracking data, and locates real-time weak nodes based on the learning situation portrait and the text response tracking data analysis; loads the real-time weak nodes and real-time education goals into the education goal association topology, and narrows and locates the education association topology nodes; extracts the associated guidance skeleton of the education association topology node to perform random guidance on the students until the associated education goal is closed.

[0010] Compared with the prior art, the beneficial effects of the technical solution of the present invention are:

[0011] The educational question-and-answer robot matches the corresponding target guidance skeleton according to the students' real-time questions, and dynamically adjusts the learning path according to the students' feedback. This process ensures the personalization of educational content and can generate customized learning paths according to the actual needs and understanding level of each student; the static guidance words bound to the starting guidance node are filled with question-and-answer scene features according to the students' real-time questions, and the starting dynamic guidance words are generated, so that each student can learn at his or her own pace and ability, avoiding a one-size-fits-all teaching method, thereby improving learning efficiency and effectiveness; the educational question-and-answer robot makes gradual adjustments based on the students' response text to the starting dynamic guidance words, so that the educational question-and-answer robot can respond to the students' learning status in real time, provide timely help or challenges, and enhance the interactivity and pertinence of the learning experience; by calling text response tracking data and learning situation portraits, the educational question-and-answer robot can accurately identify students' real-time weak nodes This process enables the educational question-answering robot to focus on guiding students' weak links, and continuously optimize teaching content during the learning process to improve the effectiveness and pertinence of learning; loading real-time weak nodes and real-time educational goals into the educational goal association topology can limit the scope of the learning path and ensure that students concentrate on solving relevant educational goals and knowledge points, which makes the selection of educational goals and teaching content more accurate and ensures that every step of students' learning is closely related to the current educational goals; by performing random guidance, the educational question-answering robot provides different learning path options, which enhances the flexibility and diversity of the learning process and helps students access and master knowledge points from multiple angles. This random guidance mechanism can avoid the simplification of learning paths, provide students with different learning methods and problem-solving ideas, enhance the fun and challenge of learning, and also help to comprehensively improve students' understanding ability. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 A flow chart of the dynamic text response method of the educational question-and-answer robot provided in an embodiment of the present application.

[0013] Figure 2 A schematic diagram of the structure of the dynamic text response system of the educational question-and-answer robot provided in an embodiment of the present application.

[0014] Explanation of the accompanying symbols: starting guidance node positioning module 10, question and answer scenario feature filling module 20, binding guidance word filling module 30, real-time weak node positioning module 40, associated topology node positioning module 50, random guidance module 60. DETAILED DESCRIPTION

[0015] In order to more clearly understand the above-mentioned objects, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that, in the absence of conflict, the embodiments of the present application and the features therein can be combined with each other.

[0016] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the scope of protection of the present invention is not limited to the specific embodiments disclosed below. Example 1

[0017] like Figure 1 As shown, this embodiment discloses a dynamic text response method of an educational question-answering robot, the method comprising:

[0018] After the educational question-answering robot matches the target guidance skeleton corresponding to the real-time educational goal according to the real-time questions asked by the students, it locates the starting guidance node in the target guidance skeleton.

[0019] The educational question-and-answer robot first receives students' real-time questions, parses the questions using NLU (natural language understanding) technology, and converts them into real-time question-answer triples, which contain information such as the subject, question type, and cognitive complexity of the student's question. Based on the parsed real-time question-answer triples, it matches the corresponding real-time educational goals. Real-time educational goals are set based on the textbook content and the student's learning progress, such as the mastery of a certain knowledge point or the ability to answer a certain type of question.

[0020] The goal guidance skeleton is a guidance structure organized according to real-time educational goals. It contains a series of guidance nodes, which can be static (such as fixed guidance language) or dynamic (such as guidance content that changes based on student responses). By matching the real-time educational goals, the starting guidance node in the goal guidance skeleton is located. The starting guidance node is the first step in the educational goal, marking the starting point of interaction with students. During this process, the educational question-and-answer robot refers to the student's learning profile to determine the most appropriate starting guidance node. If a student has weaknesses in a certain knowledge point, a simpler node is selected to start guidance, ensuring that the student can smoothly enter the learning state.

[0021] The static guiding words bound to the starting guiding node are filled with question-answering scene features according to the real-time questions asked by the students, and the starting dynamic guiding words are output.

[0022] The initial guidance node is pre-configured with static introductory text. This text is fixed text designed based on educational objectives and course content, and is intended to guide students into the learning scenario. For example, a guide for a specific knowledge point might be, "Today we're going to learn how to solve a quadratic equation." Based on students' real-time questions, the static introductory text is populated with question-and-answer scenario features. This process involves both explicit and implicit verification. Specifically, explicit verification verifies whether students have understood certain basic concepts based on keywords in their real-time questions. For example, if a student asks, "How to solve the discriminant of a quadratic equation?" this indicates that the student has already grasped some of the concepts, rather than having to start from scratch. Implicit verification infers the student's implicit knowledge needs based on their question. For example, a student's question may not directly involve the specific formula for a quadratic equation, but the wording of the question may indicate that the student is confused about the equation-solving process. In this case, the introductory text can be inferred and dynamically adjusted.

[0023] The analysis is conducted in combination with the answer time and the learning situation portrait. Specifically, if the student takes too long to respond to a certain guide, it is judged that there is a problem with the student's understanding, and the guide is dynamically adjusted, for example by providing more prompts or more intuitive explanations; the learning situation portrait is a personalized learning profile of the student constructed based on data such as the student's learning history, error frequency, and answering speed. The learning situation portrait can be used as a reference to adjust the depth and difficulty of the guide. For example, if the learning situation portrait shows that the student has a high error frequency in a certain type of math problem, more emphasis will be placed on guiding from the basics rather than jumping directly into complex problems.

[0024] According to the above process, the dynamically adjusted starting dynamic guide words are finally output to ensure that students can successfully achieve their learning goals during the interactive process.

[0025] The educational question-and-answer robot iteratively performs cross-level dynamic matching of guidance nodes and binding guidance word filling in the target guidance skeleton based on the student's response text to the initial dynamic guidance word until the real-time education goal is achieved in a closed loop.

[0026] During the interaction with the educational Q&A robot, students respond to the initial dynamic guidance and answer the robot's guiding questions. The robot's task is to continue guiding learning based on the student's responses and help the student achieve their learning goals. Based on the student's response text, the robot dynamically matches guidance nodes across levels along the target guidance skeleton and populates binding guidance words. Specifically, the target guidance skeleton is hierarchical, with each guidance node guiding the student into a different learning path. Based on the student's response text, the robot dynamically selects the appropriate next guidance node. This selection is not limited to nodes at the current level; it may also span to other levels within the target skeleton, ensuring that the robot can select the appropriate learning path based on the student's actual needs. After matching the next guidance node, the robot fills in static or dynamic guidance words based on the node's characteristics, generating guidance content appropriate for the student's current learning stage. For example, if the student answers correctly, the robot selects a further, in-depth question; if the student answers incorrectly, the robot selects simplified guidance words to help the student understand the essence of the question.

[0027] This process is cyclical. The educational question-and-answer robot continuously adjusts the guiding language according to the student's answer each time until the student achieves the real-time educational goal. This closed loop means that the student's learning goal has been effectively guided and achieved. For example, the student's learning goal is to "master the solution method of quadratic equations." The educational question-and-answer robot adjusts the guiding language according to the student's answer each time, and gradually guides the student to complete the learning of the knowledge points until the student can correctly answer all related questions and achieve the preset learning goals.

[0028] Call the text response tracking data, and locate the real-time weak nodes based on the learning situation portrait and the text response tracking data analysis.

[0029] The educational Q&A robot tracks student responses in real time, recording information such as the content of each interaction, response time, and error rate. This textual response tracking data reflects the student's understanding and mastery of the learning content. The learning profile, constructed based on data such as the student's learning history, error frequency, and answering speed, helps identify areas of weakness and key knowledge points that require further refinement.

[0030] Based on text response tracking data and learning situation portraits, real-time weak nodes in students' learning process are identified. Real-time weak nodes refer to knowledge points or links where students perform poorly in the current learning process. For example, if students make many mistakes in answering a certain type of question or do not have a thorough understanding of certain concepts, this part will be marked as a real-time weak node, and special attention will be paid to this part in the subsequent guidance, providing more detailed explanations and exercises.

[0031] The real-time weak nodes and real-time education goals are loaded into the education goal association topology, and the education association topology nodes are narrowed and located.

[0032] The educational goal association topology is a graph structure used to represent the relationships between educational goals. In the educational goal association topology, each node represents an educational goal, and the connections between nodes indicate the relationships between goals. Real-time weak nodes are knowledge points where students perform poorly during the learning process. Adding real-time weak nodes to the educational goal association topology helps identify areas where students need more help. Real-time educational goals are learning objectives set during this interaction. Adding real-time educational goals to the educational goal association topology helps the educational question-answering robot maintain goal orientation during the teaching process. The educational goal association topology continuously adjusts to the student's learning progress. Through a narrowing process, the educational question-answering robot focuses on the educational association topology nodes that are most relevant to the real-time weak nodes and real-time educational goals. This process is equivalent to focusing only on those parts of a large learning goal network that are closely related to the student's current learning status and weaknesses. This prevents students from being distracted by irrelevant goals and content, ensuring a more personalized and efficient teaching process.

[0033] The associated guidance skeleton of the education associated topological node is extracted to perform random guidance on the students until the associated education goal reaches a closed loop.

[0034] Each education-related topology node has an associated guidance skeleton, which indicates how to guide students to complete the learning task of the goal. The associated guidance skeleton contains a series of preset learning paths and methods, such as steps, questions, interactive tasks, etc., to help students gradually achieve their goals.

[0035] Random guidance means that according to the different learning paths set in the associated guidance skeleton, the educational question-and-answer robot randomly selects multiple paths for guidance, helping students gradually achieve educational goals in different learning scenarios. Random guidance can avoid the learning process being too single or mechanical, making the students' learning experience more diverse. For example, under the learning goal of "quadratic equations", the educational question-and-answer robot can randomly select different guidance paths, such as asking students about the specific method of "solving equations", or guiding students to perform actual calculation operations, and then give appropriate feedback based on the students' answers.

[0036] As students continue to participate in random guidance, the educational question-and-answer robot gradually adjusts the learning content and path based on students' feedback until students complete the learning tasks associated with the educational goals. The closed loop means that students have mastered the relevant knowledge points and can successfully solve problems related to the educational goals. The educational question-and-answer robot will confirm the students' learning progress to ensure that their learning achieves the expected goals.

[0037] After the educational question-answering robot matches a target guidance skeleton corresponding to a real-time educational goal according to a student's real-time question, the method locates a starting guidance node in the target guidance skeleton.

[0038] The educational question-and-answer robot uses NLU to parse the students' real-time questions and outputs real-time question-answer triples; matches and calls the target guidance skeleton in the static skeleton library based on the real-time question-answer triples; calls the learning situation portrait according to the student login ID; calibrates the guidance starting point in the target guidance skeleton based on the learning situation portrait, and locates the starting guidance node.

[0039] Students' real-time questions are expressed in natural language, and educational Q&A robots require NLU (natural language understanding) technology to parse them. NLU converts students' natural language questions into structured data, including subject, question type, and cognitive complexity, so that the system can understand and respond. Specifically, NLU identifies the subject information in the question. For example, if a student asks about algebra or geometry, the corresponding mathematics subject is extracted from the question. Question types, such as multiple-choice, fill-in-the-blank, and essay, are analyzed. Each question type reflects the way a knowledge point is tested, and this categorization allows for more targeted guidance. Cognitive complexity refers to the difficulty level or cognitive skills involved in the question. Based on the question format and content, the depth of the student's understanding can be estimated. For example, "How to solve a quadratic equation" is a relatively basic question, while "How to prove the discriminant of a quadratic equation" requires higher cognitive complexity. Based on the parsing results, real-time question-answer triples are output, which serve as the basis for matching the target guidance skeleton.

[0040] The static skeleton library contains various educational objectives, guidance paths, and pre-set questions. The guidance skeletons in the static skeleton library cover knowledge points and learning objectives at different levels, from basic to advanced. Real-time question-answer triples are matched within the static skeleton library. This includes matching relevant knowledge areas based on subject, matching relevant question types and problem-solving methods based on question type, and matching guidance paths appropriate to the student's current ability level based on cognitive complexity. Based on the matching results, the goal-oriented guidance skeleton that matches the student's question is called.

[0041] The learning profile is called up according to the student's login ID. The learning profile is a personalized learning profile of the student constructed based on data such as the student's learning history, error frequency, and answering speed. It is used to comprehensively display the student's learning status and provide a basis for personalized teaching.

[0042] The goal-oriented guidance framework consists of multiple guidance nodes, each corresponding to a stage of student learning. Based on the student's learning profile, the starting point for guidance is calibrated within the goal-oriented guidance framework to address the student's actual needs. For example, if a student has a strong grasp of a particular knowledge point, a more advanced guidance node is selected to begin, moving directly to the relevant, in-depth questions. If a student has a weak grasp of a particular knowledge point, a basic guidance node is selected, starting with the simplest content and gradually helping the student build a solid foundation. This process locates the starting guidance node, ensuring that guidance is tailored to the student's current learning level, thereby improving learning efficiency and effectiveness.

[0043] The educational question-and-answer robot iteratively performs cross-level dynamic matching of guidance nodes and binding guidance word filling in the target guidance skeleton based on the student's response text to the initial dynamic guidance word until the real-time educational goal is achieved in a closed loop. The method includes:

[0044] After receiving the student's initial response text in response to the initial dynamic guide, taking the initial guide node as the starting point, the initial response text is used to connect the out-degree hierarchical node along the target guide skeleton to match and locate the second guide node; the static guide bound to the second guide node is filled with question-and-answer scenario features based on the initial response text, and the second dynamic guide is output; after receiving the student's second response text in response to the second dynamic guide, taking the second guide node as the starting point, the second response text is used to connect the out-degree hierarchical node along the target guide skeleton to match and locate the third guide node; according to the student's response reply text, the cross-level jump of the guide node is iteratively executed in the target guide skeleton until the real-time education goal is achieved in a closed loop.

[0045] In the interaction of the educational question-answering robot, students respond to the initial dynamic guidance language, and the educational question-answering robot receives the student's initial response text, which reflects the student's understanding and response to the initial guidance content, including the student's answer to the question, explanation or further inquiry.

[0046] The goal guidance skeleton is a hierarchical structure. Each guidance node corresponds to a knowledge point or task in the educational goal. The starting response text is connected to the goal guidance skeleton to determine the next learning path. This connection process is cross-level, that is, the educational question-answering robot decides whether to enter the next level node in the goal guidance skeleton based on the student's response. Each guidance node has several out-degrees, that is, the next node that can be jumped to. The educational question-answering robot matches the most suitable second guidance node based on the student's response. For example, if a student accurately answers a question about a basic concept, the educational question-answering robot will guide the student to the node of the next level to continue discussing the in-depth content of the knowledge point. The goal is to ensure that the student's answer can be guided to the next appropriate node and that the learning path is coherent.

[0047] Each guide node is bound to static guides. These static guides are pre-set and designed to help students understand the current knowledge point or task. Based on the initial response text, the static guides bound to the second guide node are filled with question-and-answer scenario features. That is, the content of the static guides is adjusted according to the students' answers to make them more in line with the students' needs. This is a process of combining static guides with actual question-and-answer scenarios. Based on the students' learning status, cognitive level, and the context of the questions, the blanks in the static guides are filled in to make them personalized dynamic guides. For example, if a student has questions about a certain concept, a more detailed explanation or example is added to the guide. After the static guides are filled in, the second dynamic guide is generated. This is a guide that is dynamically generated based on the students' answers and learning progress. The second dynamic guide not only contains the preset guide content, but can also be flexibly adjusted according to the students' specific circumstances.

[0048] Students respond to the second dynamic guide and return a second response text. The second response text reflects the student's understanding, answer, or further questions about the second dynamic guide. Based on the second response text, the student continues to connect the out-degree hierarchical nodes along the learning path according to the structure of the target guidance skeleton and selects an appropriate third guidance node. The third guidance node is a higher level in the target guidance skeleton and can pose more challenging questions or tasks based on the student's depth of understanding and knowledge mastery.

[0049] Cross-level jumps are iteratively executed within the goal-guided framework. Cross-level jumps mean that the robot not only connects nodes within the current level but also jumps from one level to another based on the student's learning progress, ensuring that the student's learning path always aligns with their understanding and mastery. For example, if a student has mastered the necessary concepts at a certain level, they can be guided across levels to a more advanced knowledge point. Alternatively, if the student's understanding is still insufficient, they can return to a lower level for more basic explanations. Each student's response text influences the choice of learning path. The educational Q&A robot adjusts the jumps within the goal-guided framework based on the student's understanding, error frequency, and response accuracy.

[0050] The real-time education goal achievement closed loop refers to the students achieving the predetermined education goals during the learning process. Whenever a student successfully masters the current knowledge points through a round of iterative learning, the education question-answering robot will confirm whether the learning goal has been achieved. When all relevant guidance nodes are completed and the student can correctly answer questions or master skills, the real-time education goal achievement closed loop is determined. At this time, the learning process of the real-time education goal ends and the student has achieved the expected learning outcomes.

[0051] The method further comprises:

[0052] Based on the decomposition of the textbook catalog, multiple sample education units are obtained as multiple sample education objectives; knowledge predecessor and successor association analysis is performed on the multiple sample education units to construct multiple tree-shaped knowledge skeletons; static guide language binding of tree nodes is performed on the multiple tree-shaped knowledge skeletons to obtain multiple tree-shaped guidance skeletons; multiple historical normalized situation information of the multiple sample education units is called to construct multiple education guidance triples, wherein the education guidance triples are composed of sample subjects, sample question type sets and sample cognitive complexity; after using the multiple education guidance triples to identify the multiple sample education objectives, the multiple sample education objectives and multiple tree-shaped guidance skeletons are associated and stored to complete the construction of the static skeleton library.

[0053] The textbook catalog is a systematic framework for course content, encompassing all teaching units and their corresponding knowledge points. Based on the textbook catalog, the course content is broken down into multiple sample educational units. A sample educational unit is an independent component of the course and can include a knowledge point, a skill, or a specific task. For example, a mathematics course might include sample educational units such as "Solving Quadratic Equations" and "Calculating the Area of ​​Geometric Figures." Each sample educational unit is considered a sample educational objective. A sample educational objective might be for students to master a certain knowledge point, complete a certain task, or understand a certain method. By breaking down the textbook catalog, multiple sample educational objectives can be generated.

[0054] In education, some knowledge points serve as the foundation for others, meaning they serve as prerequisites for subsequent learning. For example, before learning "quadratic function graph analysis," students need to master the solution to "quadratic equations." Based on the textbook content, we conduct a knowledge predecessor and successor correlation analysis to identify the dependencies between various knowledge points. Based on these dependencies, we construct multiple tree-like knowledge skeletons. Each tree-like knowledge skeleton contains relevant educational objectives and knowledge points. The structure resembles a tree, with knowledge points arranged sequentially. The root node is the most basic knowledge point or task, and the child nodes represent the associated advanced knowledge or tasks. This way, when students learn a particular knowledge point, this tree structure automatically guides them to learn the prerequisite knowledge for that point and gradually transition to subsequent knowledge. By analyzing the predecessor and successor relationships between sample educational units, we connect knowledge points and educational objectives into an orderly learning chain. The connections between each node represent the knowledge dependencies, thus constructing a clear knowledge path.

[0055] Each tree node in the knowledge tree represents an educational objective or knowledge point. Each tree node is associated with a corresponding static guide, which is a standard guide text or prompt provided to students when they learn that node. The static guide is pre-set based on the content of the knowledge point and can include simple explanations, demonstrations, or question prompts to help students understand the knowledge point or task. By binding the static guide to each node in the knowledge tree to form a tree-shaped guide skeleton, each node is accompanied by the corresponding static guide, ensuring that students receive appropriate prompts and support during the learning process.

[0056] Learning information is data that records students' learning process and performance, including students' learning history, accuracy, problem-solving speed, learning progress, etc. In order to construct an education guidance triplet, it is necessary to call multiple historical normalized learning information of multiple sample education units. Normalization refers to the standardized processing of learning information of different students and converting it into a unified and comparable format. Through this process, historical normalized learning information can be compared across different students and different learning situations, so that teaching strategies can be more personalized.

[0057] The educational guidance triplet consists of sample subjects, sample question type sets and sample cognitive complexity. Among them, sample subjects refer to the subject areas involved in the sample education unit, such as mathematics, physics, Chinese, etc. Each sample education unit has a corresponding sample subject; the sample question type set is a collection of question types involved in each sample education unit, such as multiple-choice questions, fill-in-the-blank questions, and essay questions. Each question type reflects the way of examining the knowledge points of the sample education unit; the sample cognitive complexity represents the cognitive difficulty level of the sample education unit. According to methods such as Bloom's cognitive goal taxonomy theory, cognitive complexity can be divided into multiple levels from low to high, such as memory, understanding, application, analysis, synthesis, and evaluation.

[0058] By calling on historical normalized context information, corresponding educational guidance triples are constructed based on the subject, question type and cognitive complexity of each sample education unit. These educational guidance triples provide important reference for the educational question-answering robot to match the target guidance skeleton in subsequent steps.

[0059] Based on the educational guidance triples, the specific characteristics of each sample educational goal are identified and determined. Each sample educational goal has a corresponding tree-shaped guidance skeleton, which describes the student's learning path from basic to advanced. The sample educational goals and the corresponding tree-shaped guidance skeleton are associated and stored, and this information is stored in a static skeleton library. The static skeleton library contains a large number of sample educational goals and their related tree-shaped guidance skeletons. In this way, the educational question-and-answer robot can quickly find the goals and guidance paths that are most relevant to the current student's learning situation in the subsequent teaching process, thereby realizing personalized learning.

[0060] According to the real-time question-answer triples, the target guiding skeleton is matched and called in the static skeleton library, and the method includes:

[0061] Calculate the correlation between the multiple question-answer vectors of the real-time question-answer triple and the multiple education guidance triples; locate the real-time education target among the multiple sample education targets by serializing the multiple question-answer vector correlations; and call the tree-shaped guidance skeleton of the real-time education target as the target guidance skeleton.

[0062] The real-time question-and-answer triples are compared with multiple educational guidance triples, and their similarity is evaluated by calculating the correlation between their question-and-answer vectors. Specifically, each triple can be converted into a question-and-answer vector, where each dimension corresponds to a specific numerical value of the subject, question type, and cognitive complexity. By converting the numerical values ​​of these dimensions into vectors, the vector similarity between each pair of triplets is calculated, usually using cosine similarity, Euclidean distance, or other similarity measurement methods. Based on the vector similarity, it is possible to identify which educational guidance triples are most compatible with the real-time question-and-answer triples.

[0063] The calculated correlations of multiple question-answer vectors are serialized. Specifically, all question-answer vector correlations are arranged from highest to lowest, with the highest correlations prioritized. This ultimately creates a sequence where the first element represents the most optimal educational guidance triple, followed by the next best triple, and so on. This sequence is used to determine the educational guidance triple that best matches the student's current question and to determine the real-time educational goal.

[0064] Each educational goal has a corresponding tree-shaped guidance skeleton. According to the determined real-time educational goal, the corresponding tree-shaped guidance skeleton is called as the goal guidance skeleton. This tree-shaped guidance skeleton contains all relevant learning nodes and guides students to gradually master relevant knowledge through these nodes.

[0065] The real-time weak nodes and the real-time education goals are loaded into the education goal association topology, and the education association topology nodes are narrowed and located. The method includes:

[0066] Construct an education goal association matrix for the multiple sample education goals; calculate the node attribute overlap of the multiple tree-shaped knowledge skeletons, and use the calculation results to fill the education goal association matrix; binarize the education goal association matrix using a preset overlap scale, and then convert the education goal association matrix into the education goal association topology; use the goal-guiding skeleton as a matching taboo, use the real-time weak node to traverse the static skeleton library, and locate P associated education goals; load the associated education goals and real-time education goals into the education goal association topology, and limit the location of the education association topology nodes.

[0067] There may be certain dependencies or similarities between multiple sample educational objectives. In order to better identify these relationships, an educational objective association matrix is ​​constructed. The educational objective association matrix is ​​a two-dimensional matrix, in which each row and column represents a sample educational objective. Each element in the matrix represents the degree of association between two sample educational objectives, such as their similarity or dependency in the learning path.

[0068] By analyzing multiple sample educational objectives, the relationship strength between each pair of sample educational objectives is calculated. For example, some sample educational objectives are related to each other, or they have some intersection in cognitive complexity. The matrix is ​​filled in according to factors such as the similarity, dependency and difficulty level between the sample educational objectives to obtain the educational objective association matrix. For example, the element value of the matrix represents the similarity in cognitive difficulty between one sample educational objective and another sample educational objective, or whether they belong to the same knowledge field.

[0069] Each tree-shaped knowledge skeleton is a knowledge structure, which represents the various knowledge points that students need to master through hierarchical nodes. Each node has certain attributes, such as knowledge point type, cognitive complexity, learning path, etc. According to the node attributes in the tree-shaped knowledge skeleton, the node attribute overlap between different tree-shaped knowledge skeletons is calculated. Specifically, if there are multiple similar node attributes in two tree-shaped knowledge skeletons, their node attribute overlap will be higher. The calculated result of the node attribute overlap is filled in the corresponding position in the education goal association matrix. If the two tree-shaped knowledge skeletons have a high overlap in node attributes, then the association value between them will also be higher.

[0070] Binarization is the process of converting the association matrix into a simpler binary form. A preset overlap scale is set to determine whether the association between two sample educational objectives is strong enough. If the association between the two sample educational objectives is greater than the preset overlap scale, the two sample educational objectives are considered related, and their positions in the matrix are marked as 1. If the association between the two sample educational objectives is lower than the preset overlap scale, they are considered to have no strong association, and the corresponding positions in the matrix are marked as 0. After binarization, the educational objective association matrix becomes a binary matrix, where 1 indicates a relationship and 0 indicates no relationship. The resulting educational objective association topology is a graph structure used to represent the relationship between educational objectives. It can help educational question-answering robots better understand the dependencies and order between different educational objectives.

[0071] Using the goal-guided skeleton as a matching taboo means avoiding selecting inappropriate or already mastered educational objectives in the learning path. Real-time weak nodes refer to knowledge points where students demonstrate weakness or insufficient understanding during the learning process. Real-time weak nodes are used to traverse the static skeleton library to find the educational objectives most relevant to them. This process typically uses search algorithms, such as breadth-first search or depth-first search, to locate objectives that match the weak nodes. By traversing the static skeleton library, the relevance of each educational objective is checked and compared with the real-time weak nodes. P related educational objectives are selected. These related educational objectives are the ones that are most likely to help students address their weaknesses and improve their learning outcomes.

[0072] The determined associated educational goals and real-time educational goals are loaded into the educational goal associated topology. Narrowing and positioning refers to screening out the topological nodes most relevant to the current learning path by analyzing the educational goal associated topology. In this step, limiting the learning path by the loaded associated educational goals and real-time educational goals can effectively narrow the selection range of educational goals and make the learning path more accurate. Finally, through restriction and screening, the most appropriate node is located in the educational goal associated topology as the educational associated topology node to ensure that students can continue to move towards the correct learning goals.

[0073] Extracting the associated guidance skeleton of the education associated topological node and performing random guidance on students until the associated education goal reaches a closed loop, the method comprising:

[0074] According to the sample educational objectives corresponding to the educational association topological nodes, the association guidance skeleton is called from the static skeleton library; random guidance constraints are preset, wherein the random guidance constraints include cross-level guidance point selection constraints and same-level guidance point selection constraints; according to the random guidance constraints, W random guidance nodes are selected in the association guidance skeleton, and random guidance is performed on the students until the associated educational objectives are closed.

[0075] Each sample educational goal corresponds to a tree-shaped guidance skeleton. According to the sample educational goal corresponding to the education association topology node, the corresponding association guidance skeleton is called from the static skeleton library. The association guidance skeleton contains the guidance path of the sample educational goal, such as how to help students master a certain knowledge point through step-by-step guidance.

[0076] Random guidance constraints are a set of rules set for the random guidance of students' learning paths. The purpose is to provide students with diverse learning paths by randomly selecting guidance nodes and avoid a single learning method. Random guidance constraints ensure the rationality of the guidance process and prevent students from falling into irrelevant or repetitive content during the learning process. Random guidance constraints include cross-level guidance point selection constraints and same-level guidance point selection constraints. Among them, cross-level guidance point selection constraints refer to the rules set when guiding between different knowledge levels. Educational goals and knowledge points are usually organized in levels. Cross-level guidance refers to jumping from a lower-level node to a higher-level node. As learning progresses, students may need to jump from simple basic knowledge to complex applications or analysis. This cross-level guidance point selection constraint ensures that the cross-level guidance path is reasonable. For example, when jumping from basic mathematical concepts to applied mathematical problems, the robot needs to ensure that the students have mastered the basic concepts; same-level guidance point selection constraints refer to the rules when guiding within the same knowledge level. Within the same level, students may have multiple selectable learning paths or tasks. Same-level guidance constraints ensure that irrelevant or repetitive learning content will not be selected, but instead provide content that best matches the students' current progress.

[0077] Based on the random guidance constraints, W random guidance nodes are selected from the associated guidance skeleton. The W random guidance nodes are not randomly selected, but are selected under the constraints to best help students achieve their educational goals. For example, if students are learning "how to solve quadratic equations," the W random guidance nodes selected may involve different solution steps or related applications of quadratic equations.

[0078] W randomly selected guidance nodes are provided to students, and interactive guidance is used to help them complete their learning tasks. Each random guidance node corresponds to a specific learning task or question, and students need to respond and complete the relevant learning objectives. As students complete each random guidance node, the learning path is continuously adjusted based on their performance and feedback to ensure that they can successfully master the knowledge until the associated educational objectives are achieved. This means that students have successfully mastered the knowledge points and skills of the educational objectives and can correctly answer related questions or solve tasks.

[0079] Calibrate the guidance starting point on the target guidance skeleton according to the learning situation portrait and locate the starting guidance node, the method comprising:

[0080] Extract multiple target guidance nodes from the target guidance skeleton; use multiple groups of conventional error labels of the multiple target guidance nodes to traverse the learning situation portrait, and count to obtain the error frequencies of multiple stages; after weighting the error frequencies of multiple stages according to the node level characteristics, perform weighted result compensation based on the time series characteristics to obtain multiple node weight coefficients; based on the multiple node weight coefficients, perform guidance starting point calibration on the target guidance skeleton to locate the starting guidance node.

[0081] The goal-guidance skeleton is a structure composed of multiple nodes. Each node represents a specific learning task, knowledge point or educational goal. These nodes are organized according to a certain hierarchical relationship to form an orderly learning path. Multiple goal-guidance nodes are extracted from the goal-guidance skeleton. These goal-guidance nodes are the tasks or knowledge points that students need to complete during the learning process.

[0082] Each target guidance node has one or more common error labels, which identify common error types students may encounter when learning that target guidance node. These labels include misunderstandings of specific knowledge points and common deviations in problem-solving. The learning profile is a personalized data archive that records a student's learning history and performance, including their error records, learning progress, and knowledge mastery. By traversing the student's learning profile based on the common error labels and counting the frequency of errors encountered at each stage, we can obtain error frequencies for multiple stages, thereby determining which learning links students perform poorly in and which errors they are prone to making.

[0083] Each target guidance node has a different level in the target guidance skeleton. Some target guidance nodes belong to the basic level, while other target guidance nodes belong to more advanced tasks or knowledge points. A weight is assigned to each target guidance node according to the node level characteristics of the target guidance node. For example, the weight of the basic node is lower, while the weight of the advanced node is higher, because the advanced node usually has a greater impact on students' learning progress. According to the assigned weights, the error frequencies of multiple stages are weighted, which means that the nodes with higher levels and lower error frequencies will have higher weights, while the nodes with lower levels and higher error frequencies will have lower weights.

[0084] Temporal features refer to the temporal characteristics of a student's learning process, including their learning progress and the intervals between learning stages. Based on these temporal features, the weighted results are compensated. For example, if a student experiences long pauses during learning, or if the frequency of errors increases significantly at a certain stage, this data is compensated and the node weights are adjusted to ensure the rationality of the learning path. Through hierarchical weighting and temporal compensation, the node weight coefficient for each target guidance node is ultimately calculated. These node weight coefficients reflect the importance and difficulty of each node in the student's learning process, helping the robot determine which nodes require more attention and which nodes can be skipped or simplified.

[0085] Based on the obtained weight coefficients of multiple nodes, the guidance starting point is calibrated within the target guidance skeleton. The guidance starting point is the starting node in the learning process, marking where students begin learning a particular educational goal. The purpose of guidance starting point calibration is to ensure that students begin at the learning node that best suits their current state. For example, if a student makes a lot of errors on a certain knowledge point, they will choose to start at a foundational node to consolidate their basic knowledge. After calibration, the starting guidance node is located. This starting guidance node is the first task or knowledge point that students learn. Based on the student's learning profile and historical error data, it ensures a personalized learning path and helps students address weaknesses.

[0086] Calling text response tracking data and locating real-time weak nodes based on the learning situation portrait and the text response tracking data analysis, the method includes:

[0087] The text response tracking data and the learning situation portrait are compared to obtain multiple co-occurring wrong questions; based on the topic attributes of the multiple co-occurring wrong questions, cross-disciplinary association of knowledge points is performed to locate the real-time weak nodes.

[0088] Text response tracking data includes information such as each student's answer when interacting with the educational question-answering robot, the time of answering, whether it was correct or not, and the type of error. By analyzing text response tracking data, we can understand students' performance on specific questions and which questions they are prone to making mistakes. Learning profiles are personalized records of students' learning status, including their historical learning data, error frequency, knowledge points mastered, and current learning progress. By analyzing learning profiles, we can identify students' weaknesses in past learning and their performance on specific question types or knowledge points. By comparing students' text response tracking data with their learning profiles, we can identify questions that students repeatedly make mistakes on across multiple learning sessions and obtain multiple co-occurring incorrect questions.

[0089] Each co-occurring incorrect question has specific attributes, such as the subject type (e.g., mathematics, physics), question type (multiple-choice, fill-in-the-blank, essay), knowledge points (e.g., algebra, geometry, trigonometry), and cognitive difficulty. Interdisciplinary knowledge point correlation refers to the correlation between knowledge points across different subjects. This process uses question attributes to identify related problems that may exist in other subjects. This process analyzes the questions in which students make errors and identifies similar weaknesses across different subjects, known as real-time weak nodes. These real-time weak nodes refer to knowledge points or areas of ability in which students demonstrate significant weaknesses during their current learning process. These real-time weak nodes are not limited to a single subject but are common across multiple subjects. Once these real-time weak nodes are located, the learning path is adjusted accordingly, providing students with more targeted instructional content to help them address their weaknesses.

[0090] In summary, the dynamic text response method of the educational question-answering robot provided in the embodiments of the present application has the following technical effects:

[0091] The educational question-and-answer robot matches the corresponding target guidance skeleton according to the students' real-time questions, and dynamically adjusts the learning path according to the students' feedback. This process ensures the personalization of educational content and can generate customized learning paths according to the actual needs and understanding level of each student; the static guidance words bound to the starting guidance node are filled with question-and-answer scene features according to the students' real-time questions, and the starting dynamic guidance words are generated, so that each student can learn at his or her own pace and ability, avoiding a one-size-fits-all teaching method, thereby improving learning efficiency and effectiveness; the educational question-and-answer robot makes gradual adjustments based on the students' response text to the starting dynamic guidance words, so that the educational question-and-answer robot can respond to the students' learning status in real time, provide timely help or challenges, and enhance the interactivity and pertinence of the learning experience; by calling text response tracking data and learning situation portraits, the educational question-and-answer robot can accurately identify students' real-time weak nodes This process enables the educational question-answering robot to focus on guiding students' weak links, and continuously optimize teaching content during the learning process to improve the effectiveness and pertinence of learning; loading real-time weak nodes and real-time educational goals into the educational goal association topology can limit the scope of the learning path and ensure that students concentrate on solving relevant educational goals and knowledge points, which makes the selection of educational goals and teaching content more accurate and ensures that every step of students' learning is closely related to the current educational goals; by performing random guidance, the educational question-answering robot provides different learning path options, which enhances the flexibility and diversity of the learning process and helps students access and master knowledge points from multiple angles. This random guidance mechanism can avoid the simplification of learning paths, provide students with different learning methods and problem-solving ideas, enhance the fun and challenge of learning, and also help to comprehensively improve students' understanding ability. Example 2

[0092] like Figure 2 As shown, this embodiment discloses a dynamic text response system for an educational question-answering robot, the system comprising:

[0093] The starting guidance node positioning module 10 is used for the educational question-answering robot to locate the starting guidance node in the target guidance skeleton after matching the target guidance skeleton corresponding to the real-time educational goal according to the real-time questions asked by the students.

[0094] The question-answering scene feature filling module 20 is used to fill the question-answering scene feature of the static guide words bound to the starting guide node according to the real-time questions asked by the students, and output the starting dynamic guide words.

[0095] The binding guide word filling module 30 is used for the educational question-answering robot to iteratively perform cross-level dynamic matching of guide nodes and binding guide word filling in the target guide skeleton based on the student's response text to the initial dynamic guide word, until the real-time educational goal is achieved in a closed loop.

[0096] The real-time weak node location module 40 is used to call text response tracking data and locate real-time weak nodes based on the learning situation portrait and the text response tracking data analysis.

[0097] The associated topology node positioning module 50 is used to load the real-time weak nodes and real-time education goals into the education goal associated topology, and narrowly locate the education associated topology nodes.

[0098] The random guidance module 60 is used to extract the association guidance skeleton of the education association topology node and perform random guidance on the students until the association education goal reaches a closed loop.

[0099] The starting and guiding node positioning module 10 is used to perform the following operation steps:

[0100] The educational question-and-answer robot uses NLU to parse the students' real-time questions and outputs real-time question-answer triples; matches and calls the target guidance skeleton in the static skeleton library based on the real-time question-answer triples; calls the learning situation portrait according to the student login ID; calibrates the guidance starting point in the target guidance skeleton based on the learning situation portrait, and locates the starting guidance node.

[0101] The binding guide filling module 30 is used to perform the following operation steps:

[0102] After receiving the student's initial response text in response to the initial dynamic guide, taking the initial guide node as the starting point, the initial response text is used to connect the out-degree hierarchical node along the target guide skeleton to match and locate the second guide node; the static guide bound to the second guide node is filled with question-and-answer scenario features based on the initial response text, and the second dynamic guide is output; after receiving the student's second response text in response to the second dynamic guide, taking the second guide node as the starting point, the second response text is used to connect the out-degree hierarchical node along the target guide skeleton to match and locate the third guide node; according to the student's response reply text, the cross-level jump of the guide node is iteratively executed in the target guide skeleton until the real-time education goal is achieved in a closed loop.

[0103] The starting and guiding node positioning module 10 is used to perform the following operation steps:

[0104] Based on the decomposition of the textbook catalog, multiple sample education units are obtained as multiple sample education objectives; knowledge predecessor and successor association analysis is performed on the multiple sample education units to construct multiple tree-shaped knowledge skeletons; static guide language binding of tree nodes is performed on the multiple tree-shaped knowledge skeletons to obtain multiple tree-shaped guidance skeletons; multiple historical normalized situation information of the multiple sample education units is called to construct multiple education guidance triples, wherein the education guidance triples are composed of sample subjects, sample question type sets and sample cognitive complexity; after using the multiple education guidance triples to identify the multiple sample education objectives, the multiple sample education objectives and multiple tree-shaped guidance skeletons are associated and stored to complete the construction of the static skeleton library.

[0105] The starting and guiding node positioning module 10 is used to perform the following operation steps:

[0106] Calculate the correlation between the multiple question-answer vectors of the real-time question-answer triple and the multiple education guidance triples; locate the real-time education target among the multiple sample education targets by serializing the multiple question-answer vector correlations; and call the tree-shaped guidance skeleton of the real-time education target as the target guidance skeleton.

[0107] The associated topology node positioning module 50 is used to perform the following operation steps:

[0108] Construct an education goal association matrix for the multiple sample education goals; calculate the node attribute overlap of the multiple tree-shaped knowledge skeletons, and use the calculation results to fill the education goal association matrix; binarize the education goal association matrix using a preset overlap scale, and then convert the education goal association matrix into the education goal association topology; use the goal-guiding skeleton as a matching taboo, use the real-time weak node to traverse the static skeleton library, and locate P associated education goals; load the associated education goals and real-time education goals into the education goal association topology, and limit the location of the education association topology nodes.

[0109] The random guidance module 60 is used to perform the following operation steps:

[0110] According to the sample educational objectives corresponding to the educational association topological nodes, the association guidance skeleton is called from the static skeleton library; random guidance constraints are preset, wherein the random guidance constraints include cross-level guidance point selection constraints and same-level guidance point selection constraints; according to the random guidance constraints, W random guidance nodes are selected in the association guidance skeleton, and random guidance is performed on the students until the associated educational objectives are closed.

[0111] The starting and guiding node positioning module 10 is used to perform the following operation steps:

[0112] Extract multiple target guidance nodes from the target guidance skeleton; use multiple groups of conventional error labels of the multiple target guidance nodes to traverse the learning situation portrait, and count to obtain the error frequencies of multiple stages; after weighting the error frequencies of multiple stages according to the node level characteristics, perform weighted result compensation based on the time series characteristics to obtain multiple node weight coefficients; based on the multiple node weight coefficients, perform guidance starting point calibration on the target guidance skeleton to locate the starting guidance node.

[0113] The real-time weak node location module 40 is used to perform the following operation steps:

[0114] The text response tracking data and the learning situation portrait are compared to obtain multiple co-occurring wrong questions; based on the topic attributes of the multiple co-occurring wrong questions, cross-disciplinary association of knowledge points is performed to locate the real-time weak nodes.

[0115] Through the above detailed description of the dynamic text response method of the educational question-answering robot in this specification, those skilled in the art can clearly understand the dynamic text response system of the educational question-answering robot in this embodiment. Since it corresponds to the method disclosed in the embodiment, the description is relatively simple. For relevant matters, please refer to the method part.

[0116] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present application. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application is not limited to the embodiments shown herein, but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein. Example 3

[0117] This embodiment discloses a readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, any step of Embodiment 1 is implemented.

[0118] In the several embodiments provided in this application, it should be understood that the disclosed methods, systems and storage media can be implemented in other ways. The system embodiments described above are merely schematic. For example, the division of the modules is merely a logical function division. In actual implementation, there may be other division methods, such as: multiple modules or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0119] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units; they may be located in one place or distributed across multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the scheme of this embodiment.

[0120] In addition, all functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may be separately used as a unit, or two or more units may be integrated into one unit; the above-mentioned integrated units may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0121] Those skilled in the art will appreciate that all or part of the steps of the above-mentioned method embodiments may be implemented by hardware associated with program instructions, and the aforementioned program may be stored in a computer-readable storage medium. When the program is executed, the program executes the steps of the above-mentioned method embodiments. The aforementioned storage medium includes various media that can store program codes, such as mobile storage devices, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical disks.

[0122] Alternatively, if the integrated units described above are implemented as software modules and sold or used as standalone products, they can also be stored on a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of the present invention, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product, stored on a storage medium, includes instructions for enabling a computer device (such as a personal computer, server, or network device) to execute all or part of the methods described in various embodiments of the present invention. The aforementioned storage media include various media capable of storing program code, such as removable storage devices, ROM, RAM, magnetic disks, or optical disks.

Claims

1. A dynamic text response method for an educational question-answering robot, characterized in that: The method comprises: After matching the target guidance skeleton corresponding to the real-time educational goal according to the real-time question asked by the student, the educational question-answering robot locates the starting guidance node in the target guidance skeleton; Filling the static guiding words bound to the starting guiding node with question-answering scene features according to the real-time questions asked by the students, and outputting the starting dynamic guiding words; The educational question-answering robot iteratively performs cross-level dynamic matching of guidance nodes and binding guidance word filling in the target guidance skeleton based on the student's response text to the initial dynamic guidance word until the real-time education goal is achieved in a closed loop; Calling text response tracking data, and locating real-time weak points based on the learning situation portrait and the text response tracking data analysis; Loading the real-time weak nodes and real-time education goals into the education goal association topology, and narrowing down and locating the education association topology nodes; The associated guidance skeleton of the education associated topological node is extracted to perform random guidance on the students until the associated education goal reaches a closed loop.

2. The dynamic text response method of the educational question-answering robot according to claim 1, characterized in that: After the educational question-answering robot matches a target guidance skeleton corresponding to a real-time educational goal according to a student's real-time question, the method locates a starting guidance node in the target guidance skeleton. The educational question-answering robot uses NLU to parse the students' real-time questions and outputs real-time question-answer triples; Matching and calling the target guide skeleton in the static skeleton library according to the real-time question and answer triples; Calling the learning profile according to the student login ID; The guidance starting point is calibrated on the target guidance skeleton according to the learning situation portrait, and the starting guidance node is located.

3. The dynamic text response method of the educational question-answering robot according to claim 2, characterized in that: The educational question-and-answer robot iteratively performs cross-level dynamic matching of guidance nodes and binding guidance word filling in the target guidance skeleton based on the student's response text to the initial dynamic guidance word until the real-time educational goal is achieved in a closed loop. The method includes: After receiving the student's initial response text in response to the initial dynamic guide, taking the initial guide node as the starting point, using the initial response text to connect along the target guide skeleton to match the out-degree hierarchical nodes and locate the second guide node; Filling the static introductory language bound to the second introductory node with question-answering scenario features according to the initial response text, and outputting a second dynamic introductory language; After receiving the second response text of the student in response to the second dynamic guide, starting from the second guide node, using the second response text to connect along the target guide skeleton to match the out-degree hierarchical nodes and locate the third guide node; According to the student's response reply text, the target guidance skeleton iteratively executes the cross-level jump of the guidance node until the real-time education goal is achieved in a closed loop.

4. The dynamic text response method of the educational question-answering robot according to claim 2, characterized in that: The method further comprises: Based on the decomposition of the textbook catalog, multiple sample education units are obtained as multiple sample education objectives; Performing knowledge predecessor and successor association analysis on the multiple sample education units to construct multiple tree-shaped knowledge skeletons; Performing static guide word binding of tree nodes on the plurality of tree-shaped knowledge skeletons to obtain a plurality of tree-shaped guide skeletons; Calling multiple historical normalized situation information of the multiple sample education units to construct multiple education guidance triples, wherein the education guidance triples are composed of sample subjects, sample question type sets, and sample cognitive complexity; After the plurality of sample educational objectives are identified by using the plurality of educational guidance triples, the plurality of sample educational objectives and the plurality of tree-shaped guidance skeletons are associated and stored to complete the construction of a static skeleton library.

5. The dynamic text response method of the educational question-answering robot according to claim 4, characterized in that: According to the real-time question-answer triples, the target guiding skeleton is matched and called in the static skeleton library, and the method includes: Calculating correlations between the real-time question-answer triple and multiple question-answer vectors of multiple education guidance triples; Positioning the real-time educational goal among the multiple sample educational goals by serializing the correlation degrees of the multiple question-answer vectors; The tree-shaped guidance skeleton of the real-time education target is called as the target guidance skeleton.

6. The dynamic text response method of the educational question-answering robot according to claim 4, characterized in that: The real-time weak nodes and the real-time education goals are loaded into the education goal association topology, and the education association topology nodes are narrowed and located. The method includes: constructing an educational goal association matrix of the plurality of sample educational goals; After calculating the node attribute overlap of the plurality of tree-shaped knowledge skeletons, the calculation results are used to fill the educational goal association matrix; After binarizing the educational goal association matrix using a preset coincidence scale, the educational goal association matrix is ​​converted into the educational goal association topology; Using the target-guided skeleton as a matching taboo, traversing the static skeleton library using the real-time weak nodes, and locating P associated educational targets; The associated educational objectives and real-time educational objectives are loaded into the educational objective association topology, and the education association topology nodes are narrowed and located.

7. The dynamic text response method of the educational question-answering robot according to claim 6, characterized in that: Extracting the associated guidance skeleton of the education associated topological node and performing random guidance on students until the associated education goal reaches a closed loop, the method comprising: According to the sample education goal corresponding to the education association topology node, calling the association guidance skeleton from the static skeleton library; Presetting random guidance constraints, wherein the random guidance constraints include cross-level guidance point selection constraints and same-level guidance point selection constraints; According to the random guidance constraints, W random guidance nodes are selected in the associated guidance skeleton, and random guidance is performed on the students until the associated educational goal reaches a closed loop.

8. The dynamic text response method of the educational question-answering robot according to claim 2, characterized in that: Calibrate the guidance starting point on the target guidance skeleton according to the learning situation portrait and locate the starting guidance node, the method comprising: Extracting multiple target guidance nodes from the target guidance skeleton; Using the multiple sets of conventional error labels of the multiple target guidance nodes to traverse the learning situation portrait, and counting to obtain the error frequencies of multiple stages; After weighting the error frequencies of the multiple stages according to the node level characteristics, weighted results are compensated based on the time series characteristics to obtain multiple node weight coefficients; According to the multiple node weight coefficients, a guidance starting point calibration is performed on the target guidance skeleton to locate the starting guidance node.

9. The dynamic text response method of the educational question-answering robot according to claim 8, characterized in that: Calling text response tracking data and locating real-time weak nodes based on the learning situation portrait and the text response tracking data analysis, the method includes: Comparing the text response tracking data with the learning situation portrait to obtain multiple co-occurring wrong questions; According to the attributes of the multiple co-occurring wrong questions, cross-disciplinary association of knowledge points is performed to locate the real-time weak nodes.

10. A dynamic text response system for an educational question-answering robot, characterized in that: A dynamic text response method for implementing the educational question-answering robot according to any one of claims 1 to 9, the system comprising: A starting guidance node positioning module is used for the educational question-answering robot to locate the starting guidance node on the target guidance skeleton after matching the target guidance skeleton corresponding to the real-time educational goal according to the real-time question asked by the student; A question-answering scene feature filling module is used to fill the question-answering scene feature of the static guide words bound to the starting guide node according to the real-time questions asked by the students, and output the starting dynamic guide words; A binding guide word filling module is used for the educational question-answering robot to iteratively perform cross-level dynamic matching of guide nodes and binding guide word filling in the target guide skeleton based on the student's response text to the initial dynamic guide word, until the real-time educational goal is achieved in a closed loop; A real-time weak node location module is used to call text response tracking data and locate real-time weak nodes based on the learning situation portrait and the text response tracking data analysis; An associated topology node positioning module, used for loading the real-time weak nodes and real-time education targets into the education target associated topology, and narrowing and positioning the education associated topology nodes; The random guidance module is used to extract the association guidance skeleton of the education association topology node and perform random guidance on students until the associated education goal reaches a closed loop.

Images