Teaching information processing method and apparatus, electronic device, and computer storage medium

Abstract

The present disclosure provides a teaching information processing method and device, and relates to the technical field of artificial intelligence. The specific implementation scheme is: receiving user input information; performing semantic and intention recognition on the user input information to obtain a cognitive state of the user on a related node in a teaching knowledge graph; updating historical dialogue information and mastery information of a target concept node in a user dialogue portrait of the user based on the cognitive state and the user input information, and obtaining a current focus node from the user dialogue portrait; selecting a target node related to the current focus node in the teaching knowledge graph based on the mastery information, and replacing the current focus node with the target node; generating a content constraint set based on the target node and the teaching knowledge graph; generating next teaching content information based on the content constraint set and the historical dialogue information, taking the next teaching content information as current teaching content information, and sending the current teaching content information.

Description

Teaching information processing methods and devices, electronic devices, and computer storage media Technical Field

[0001] This disclosure belongs to the field of artificial intelligence technology, specifically relating to natural language generation, natural language processing, and large model technology. In particular, it discloses a teaching information processing method and apparatus, electronic device, and computer-readable storage medium. Background Technology

[0002] In modern educational philosophy, personalized learning is widely recognized as key to improving teaching quality and efficiency. One of its ultimate goals is to provide each learner with a full-time, all-around private tutor. This tutor must not only be knowledgeable but also skilled in heuristic teaching, particularly the Socratic method, which guides students to think independently, discover problems on their own, and construct a knowledge system through a series of logically structured questions. This method is widely recognized as the best way to cultivate students' critical thinking and problem-solving abilities.

[0003] With the advancement of artificial intelligence, particularly large language models (LLMs), achieving the aforementioned goals seems within reach. However, when applying existing technologies directly to serious Socratic teaching scenarios, a profound and insurmountable technological gap has been identified.

[0004] For open-ended dialogue systems based on general large-scale language models (MLMs), the powerful natural language understanding and generation capabilities of LLMs can enable extremely fluent, natural, and expressive dialogues with users. However, they suffer from goal drift and logical instability, lacking a clear and persistent "teaching objective" as guidance. In multi-turn dialogues, the system is easily swayed by a student's keyword or secondary question, causing the dialogue to deviate from the pre-set teaching trajectory. LLMs do not perform fact-checking when generating content; instead, they create the most probabilistically plausible text. In education, a field with zero tolerance for the accuracy of knowledge, the factual illusions produced by LLMs are unacceptable. The essence of Socratic teaching lies in guiding, not telling. However, after being trained on massive amounts of text, LLMs inherently prefer to provide users with the most direct and comprehensive answers, thus depriving students of valuable thinking and exploration processes. Summary of the Invention

[0005] This disclosure provides a teaching information processing method and apparatus, an electronic device, and a computer-readable storage medium.

[0006] According to the first aspect, a teaching information processing method is provided, comprising: receiving user input information in which the user provides feedback on the current teaching content information; performing semantic and intent recognition on the user input information to obtain the user's cognitive state of relevant nodes in a teaching knowledge graph; the teaching knowledge graph is a graph representing the relationship between teaching concept nodes and teaching process nodes; based on the cognitive state and user input information, updating the historical dialogue information and mastery information of target concept nodes recorded in the user's user dialogue profile, and obtaining the current focus node from the user dialogue profile; based on the mastery information, selecting a target node related to the current focus node in the teaching knowledge graph, and using the target node as the current focus node in the user dialogue profile; generating a content constraint set for content generation based on the target node and the teaching knowledge graph; generating the next teaching content information based on the content constraint set and historical dialogue information, using the next teaching content information as the current teaching content information, and sending the current teaching content information to the user.

[0007] According to a second aspect, a teaching information processing device is provided, comprising: a receiving unit configured to receive user input information from a user providing feedback on current teaching content information; an identification unit configured to perform semantic and intent recognition on the user input information to obtain the user's cognitive state regarding relevant nodes in a teaching knowledge graph; the teaching knowledge graph being a graph representing the relationship between teaching concept nodes and teaching process nodes; an updating unit configured to update historical dialogue information and mastery information of target concept nodes recorded in the user's user dialogue profile based on the cognitive state and user input information, and to obtain the current focus node from the user dialogue profile; a selection unit configured to select a target node related to the current focus node in the teaching knowledge graph based on the mastery information, and to use the target node as the current focus node in the user dialogue profile; a generation unit configured to generate a content constraint set for content generation based on the target node and the teaching knowledge graph; and a sending unit configured to generate next teaching content information based on the target node and the content constraint set, to use the next teaching content information as the current teaching content information, and to send the current teaching content information to the user.

[0008] According to a third aspect, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform a method as described in any implementation of the first aspect.

[0009] According to a fourth aspect, a non-transitory computer-readable storage medium is provided that stores computer instructions for causing a computer to perform the method described in any implementation of the first aspect.

[0010] The teaching information processing method and apparatus provided in the embodiments of this disclosure firstly receive user input information from the user providing feedback on the current teaching content information; secondly, perform semantic and intent recognition on the user input information to obtain the user's cognitive state regarding relevant nodes in the teaching knowledge graph; the teaching knowledge graph is a graph representing the relationship between teaching concept nodes and teaching process nodes; thirdly, based on the cognitive state and user input information, update the historical dialogue information and mastery information of target concept nodes recorded in the user's user dialogue profile, and obtain the current focus node from the user dialogue profile; next, based on the mastery information, select a target node related to the current focus node in the teaching knowledge graph, and use the target node as the current focus node in the user dialogue profile; then, based on the target node and the teaching knowledge graph, generate a content constraint set for content generation; finally, based on the content constraint set and historical dialogue information, generate the next teaching content information, use the next teaching content information as the current teaching content information, and send the current teaching content information to the user. This enables an adaptive personalized teaching loop based on knowledge graphs, dynamically selecting and switching focus nodes according to user understanding and generating constrained next steps of content, thereby continuously pushing teaching information that is more relevant to the user's understanding and improving learning efficiency and effectiveness.

[0011] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0012] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein:

[0013] Figure 1 is a flowchart of an embodiment of the teaching information processing method according to the present disclosure;

[0014] Figure 2 is a schematic diagram of the overall architecture of the system corresponding to the open teaching information processing;

[0015] Figure 3 is a schematic diagram of the structure of an embodiment of the teaching information processing device disclosed herein;

[0016] Figure 4 is a block diagram of an electronic device used to implement the teaching information processing method of the present disclosure. Detailed Implementation

[0017] Unless otherwise expressly stated, throughout the specification and claims, the term "comprising" or its variations such as "including" or "comprises" shall be understood to include the stated elements or components without excluding other elements or other components.

[0018] The technical solutions of this disclosure are illustrated below through specific embodiments. It should be understood that one or more steps mentioned in this disclosure do not preclude the existence of other methods and steps before or after the combined steps, or that other methods and steps may be inserted between these explicitly mentioned steps. It should also be understood that these examples are for illustrative purposes only and are not intended to limit the scope of this disclosure. Unless otherwise stated, the numbering of each method step is only for the purpose of identifying each method step, and not to limit the order of each method or to limit the scope of implementation of this disclosure. Changes or adjustments to their relative relationships, without substantial changes to the technical content, can also be considered as within the scope of implementation of this disclosure.

[0019] The raw materials and instruments used in the examples are not subject to any specific restrictions on their source; they can be purchased from the market or prepared according to conventional methods known to those skilled in the art.

[0020] To address the shortcomings of existing large-scale language models, such as target drift, unreliable facts, and insufficient heuristics, as well as the problems of rigid interaction and knowledge engineering bottlenecks in traditional intelligent teaching systems, the teaching information processing method provided in this disclosure utilizes a structured teaching knowledge graph designed specifically for teaching as a real-time cognitive navigation system for large language models during dialogue generation. By applying dynamic constraints and guidance through a content constraint set during the generation process, a novel teaching dialogue generation method based on teaching knowledge graph navigation and constraints is created. This achieves a truly scalable AI Socratic tutor, improving the efficiency and effectiveness of teaching information generation. Figure 1 illustrates a flowchart 100 of an embodiment of the teaching information processing method according to this disclosure, which includes the following steps:

[0021] Step 101: Receive user input information for feedback on the current teaching content.

[0022] In this embodiment, the current teaching content information is content used to provide educational information to the user. This information can be educational information determined based on a preset learning syllabus and information from the user's user dialogue profile, meeting the teaching objectives. The current teaching content information may include the core knowledge concepts corresponding to the teaching concept nodes in the knowledge graph. For example, if the current teaching content information is "Current, voltage, and resistance, what kind of hidden relationship do you feel exists between them?", its core knowledge concepts are: current, voltage, and resistance.

[0023] In this embodiment, the user input information is multimodal information, such as text, images, and voice, fed back by the user after analyzing the teaching content information. The user input information includes a user identifier to distinguish the user, courseware information, and response information. It should be noted that the response information can be consistent with the teaching objectives of the current teaching content information; this response information can be correct or incorrect, such as the response "I guess the higher the voltage, the higher the resistance"; the response information can also be irrelevant to the teaching objectives of the current teaching content information, such as the response "I don't know the relationship between the three."

[0024] In this embodiment, the execution entity running on the teaching information processing method presents the current teaching content information while providing embedded feedback components (such as rating, options, multi-line text and attachments). When the user submits on the embedded feedback component, the client completes form validation and throttling, collects and attaches contextual metadata such as content ID, lesson number, user identifier or anonymous identifier, timestamp, device and version, etc., and obtains the user input information after de-identification and encryption, and submits it to the execution entity.

[0025] Step 102: Perform semantic and intent recognition on the user input information to obtain the user's cognitive state regarding the relevant nodes in the teaching knowledge graph.

[0026] In this embodiment, the instructional knowledge graph is a graph representing the relationships between instructional concept nodes and instructional process nodes. The instructional knowledge graph is a specially designed knowledge graph oriented towards the instructional process, and its structured design can serve the user's cognitive navigation. Specifically, each node (including instructional concept nodes and instructional process nodes) and the edges between nodes in the instructional knowledge graph have corresponding attributes, and the definitions of each node type are shown in Table 1.

[0027] Table 1

[0028] Node Type Description Core Function Concept Core Knowledge Concept The subject and goal of teaching Prerequisite Prerequisite Dependency Knowledge Used to diagnose learning obstacles and trace the path Misconception Common misconceptions or easily confused points Used to accurately identify user errors and provide targeted correction Question Heuristic questions used for guidance and exploration The core driving force of Socratic dialogue Example Specific instances or application scenarios Used to concretize abstract concepts and aid understanding Analogy Knowledge points used for analogical explanations Provide another cognitive perspective and lower the threshold for understanding surface

[0029] In Table 1, Concept belongs to the teaching concept node, while Misconception, Prerequisite, Question, Example, and Analogy all belong to the teaching process node. Based on the attributes of each node in the teaching knowledge graph and the user's mastery information of the target concept node, in each dialogue round, teaching concept or teaching process node that meets the teaching objective can be selected sequentially from the teaching knowledge graph as the target node. All target nodes in all dialogue rounds are then connected to ultimately achieve the node path for realizing the teaching objective. Among them, the target concept node is the teaching concept node that is only relevant to the user. The target concept node will be different for different users. At different learning stages of the user, the mastery information of the target concept node is different. It should be noted that the system sets up a corresponding user dialogue profile for each user. The target concept nodes in the user dialogue profile are teaching concept nodes that the system preloads or initializes according to the user's learning outline and that need to be learned.

[0030] For example, in an embodiment of Ohm's Law, when junior high school student A logs in, the system loads A's user profile. This profile already contains all the teaching concept nodes required for the junior high school electricity unit, such as current, voltage, resistance, and Ohm's Law. These teaching concept nodes are the target concept nodes.

[0031] In this embodiment, the attributes of nodes in the teaching knowledge graph include: node ID, node type (such as Concept, Prerequisite, Misconception, Question, Example, Analogy in Table 1), tag, and metadata. The attributes of the edges of the nodes define the teaching logic relationships between nodes. Specifically, the attributes of the edges include: edge ID, source node, destination node, edge type, and weight. The source node is the starting node of the edge, and the destination node is the ending node of the edge. The edge types include: hasPrerequisite, leadsTo, triggers, isExampleOf, and isAnalogyOf. Among them, hasPrerequisite indicates a prerequisite relationship, which is knowledge that must be mastered before learning a certain knowledge; leadsTo indicates a leading relationship, which indicates that advanced knowledge can be learned after learning a certain knowledge; triggers indicate that a certain knowledge or behavior will trigger another knowledge or behavior; isExampleOf indicates an example of a certain instance, which indicates that a certain instance is a concrete example of a certain concept; isAnalogyOf indicates whose analogy, which indicates that a certain analogy is used to help understand a certain abstract concept.

[0032] In this embodiment, the instructional knowledge graph not only includes instructional concept nodes, but also innovatively introduces instructional process nodes such as Misconception, Question, Analogy, and Prerequisite. The edges (relationships between nodes) in the instructional knowledge graph are not merely factual relationships like is-a or part-of, but also define instructional logical relationships such as leadsTo, triggers for error correction, and isExampleOf. Therefore, the instructional knowledge graph is not a static knowledge database, but a dynamic and executable map of instructional strategies, which is the fundamental prerequisite for achieving controllable Socratic dialogue.

[0033] In this embodiment, a natural language understanding model can be used to perform semantic and intent recognition on user input information. The natural language understanding model is used for intent recognition, semantic relevance judgment, and correctness estimation. Through the natural language understanding model, the teaching concept nodes and teaching process nodes included in the input information, as well as the user's cognitive state regarding these teaching concept nodes and teaching process nodes, can be determined. That is, relevant nodes are the teaching concept nodes and teaching process nodes in the teaching knowledge graph that are relevant to the user. Cognitive state includes: the degree of understanding of relevant nodes or the correctness of the answer.

[0034] In this embodiment, user input information is first converted (text / speech transcription) into text information. Then, the text information is normalized, segmented, and disambiguated. A pre-trained language model is used to generate a contextual semantic vector. Subsequently, a multi-label intent recognition model is used to determine the user's intent category (such as querying concepts, solving steps, error analysis, or feedback on confusion) and its confidence level. Domain named entity recognition and entity linking are performed in parallel. The teaching concept phrases and teaching process phrases in the input are aligned to "teaching concept nodes" and "teaching process nodes" in the teaching knowledge graph based on vector retrieval and a thesaurus. During node alignment... Subsequently, the correctness of user statements or answers, key step coverage, and causal logic consistency are scored using rules and discriminant models to obtain the initial comprehension level of the aligned nodes. Then, based on the concept-process-sequence / dependency equilateral type in the knowledge graph, local propagation and consistency verification are performed using graph embedding or graph neural networks. By integrating semantic similarity, intent weight, correctness score, and historical learning profiles, the cognitive state (mastery / partial mastery / not mastery) and confidence level of each relevant node are generated, and supporting evidence fragments and timestamps are output. Finally, this result is written into the user profile to drive subsequent personalized teaching and resource recommendations. The relevant nodes are teaching concept nodes and teaching process nodes in the teaching knowledge graph that are related to the user's input information.

[0035] Step 103: Based on the cognitive state and user input information, update the historical dialogue information and mastery information of the target concept node recorded in the user's dialogue profile, and obtain the current focus node from the user dialogue profile.

[0036] In this embodiment, the executing entity sets up a user dialogue profile for each user, which records all the user's historical dialogue information. Historical dialogue information is the main historical content of the interaction between the user and the executing entity in different dialogue rounds. Specifically, historical dialogue information includes: dialogue round, dialogue object, dialogue content text, etc., where the dialogue object can be the executing entity or the user, and the dialogue content text can be the information after extracting key information (such as keywords) from the user input information of the dialogue object.

[0037] In this embodiment, the user dialogue profile records the user's historical dialogue information, the target concept node in the user's relevant teaching knowledge graph, the mastery information of the target concept node, and the current focus node.

[0038] In this embodiment, mastery information refers to information used to assess the learner's level of understanding of the knowledge corresponding to the taught concept node. This mastery information can be identified through a pre-trained user cognitive state model, and the mastery information for the taught concept node is explicitly defined in the data structure of the user cognitive state model.

[0039] In this embodiment, teaching process nodes are means or tools used to teach concepts. They are not objects that users need to master. Therefore, the execution entity of the teaching information processing method will not evaluate whether the user has mastered a teaching process node. Instead, it will take the teaching process node as the current focus node and analyze the current focus node to help the user improve their mastery of the teaching concept node.

[0040] In this embodiment, the mastery information includes: mastery type and mastery level value. The mastery type, status, is an enumerated type, such as mastered, learning, unknown, confused. The mastery level value, mastery, is a floating-point number between these values ​​and is updated through a cognitive diagnostic model (such as DINA, IRT) or a Bayesian network based on interactive behavior.

[0041] In this embodiment, the process of obtaining mastery information is as follows: the observed response evidence is mapped to a potential mastery state (discrete skill mastery or continuous ability) using an appropriate latent variable model, and then the posterior probability or estimate at the individual level is output. Based on the two types of evidence, "correctness of response" (binary classification) or "degree of understanding" (multi-level / continuous rating), mastery information can be inferred using cognitive diagnostic models (represented by DINA / G-DINA) or IRT (item response theory). In this embodiment, the executing entity generates a timestamped interaction record based on the cognitive state (including intent, slot, topic / knowledge point identifier, and confidence level) output by the natural language understanding module and the original user input, and writes it into the historical dialogue information of the user's dialogue profile. It locates the relevant nodes involved in this round through entity-knowledge point mapping or knowledge graph, and calculates the evidence value e based on interaction effectiveness signals (such as whether the answer is correct / completed, the number of system error corrections, dwell time, confidence level, etc.). The mastery information of the node is updated according to the incremental rule mastery_new = (1-α)·mastery_old +α·e, while also updating the exposure count and recent interaction time. If necessary, soft propagation is performed on adjacent nodes using a decay coefficient β. Subsequently, a focus score is calculated for the candidate node: focus = w1·recency + w2·frequency + w3·intent_conf + ... w4·context_match (recency decays over time, frequency is the frequency of near-window interactions, context_match is a measure of context continuity, intent_conf is the intent content) selects the node with the highest score that exceeds the threshold τ as the current focus node. If no node meets the condition, it falls back to the previous focus node or the default root node. Finally, the updated history, information held by each relevant node, and the current focus node are persisted to the user dialogue profile for use in subsequent rounds of decision-making and personalized guidance.

[0042] Step 104: Based on the information acquired, select the target node related to the current focus node in the teaching knowledge graph, and use the target node as the current focus node in the user dialogue profile.

[0043] In this embodiment, the user's mastery of information M (such as answer accuracy, time taken, number of retries, etc.) is compared with a preset strategy threshold: if M is lower than the lower limit T_low of the preset strategy threshold, the "prerequisite supplementation" strategy is used to retrieve its predecessor dependency nodes along the prerequisite / depends_on relationship in the teaching knowledge graph to obtain candidate nodes; if M is in the middle range, the "horizontal consolidation" strategy is used to retrieve parallel nodes of the same level and topic to obtain candidate nodes; if M is higher than the upper limit T_high of the preset strategy threshold, the "advanced expansion" strategy is used to retrieve successor / advanced relationship nodes to obtain candidate nodes. For candidate nodes, weighted scoring and ranking are performed based on indicators such as semantic relevance, difficulty matching, learning diversity, and novelty, and the candidate node with the highest score is selected as the target node; if there are insufficient candidates, the user reverts to the default review node. Finally, the target node is written into the user's dialogue profile, replacing the "current focus node" field, and the selection criteria and timestamp are recorded, triggering corresponding content push and dialogue guidance to achieve dynamic focus shift based on mastery of information.

[0044] Step 105: Based on the target node and the teaching knowledge graph, generate a set of content constraints for content generation.

[0045] In this embodiment, the content constraint set is used to generate the next teaching content information. When generating the next teaching content information, the content constraint set can provide directional support for the next teaching content information. As shown in Figure 2, the dialogue strategy manager, as the decision-making center of the system corresponding to the teaching information processing method, has the core task of executing steps 101-105, achieving a bridging role. That is, the dialogue strategy manager parses the user's current input information, updates the user's dialogue profile, and then plans the best teaching intention for the next round of dialogue through the teaching knowledge graph, realizing the output of the content constraint set.

[0046] In this embodiment, firstly, an r-hop neighborhood traversal is performed in the teaching knowledge graph according to the edge type and weight of the target node to calculate the saliency and reachability of related nodes, forming a set of concepts that should be covered / optional / need to be excluded and a set of misconceptions to avoid. Then, combined with the erroneous cognitions or easily confused points related to the target node in the teaching knowledge graph, structural order constraints, question types and proportions, and the proportion of cognitive levels (such as memory / understanding / application / analysis) are generated according to the path of "prerequisites → target core → method steps → application examples → consolidation exercises → formative assessment". Cross-domain offsets, knowledge conflicts and risks of exceeding the scope of the curriculum are resolved through rules and scoring functions. Finally, the coverage, prohibited points, key terminology list, typical example templates, format and length, media proportion, prompt tone, citation norms, etc. are quantified into machine-readable content constraints (including must / should / not, threshold and template slots), and the output is the content constraint set directly consumed by the content generation engine.

[0047] Step 106: Based on the content constraint set and historical dialogue information, generate the next teaching content information, use the next teaching content information as the current teaching content information, and send the current teaching content information to the user.

[0048] In this embodiment, the content constraint set (such as required knowledge points, difficulty level, time limit, presentation format, prerequisites, etc.) is first parsed. Personalized weights are calculated by combining user profiles and learning records. Candidate resources are retrieved and sorted in the content library through a rule engine and recommendation model to determine teaching units and pace. Then, "next teaching content information" is assembled and generated, which includes structured fields such as topic, learning objectives, key knowledge points, estimated duration, resource list, activities / exercises and assessment methods, etc., and is assigned a unique ID and effective time. This information is written into the user's course progress and overwritten as "current teaching content information", and the current teaching content information is pushed to the user.

[0049] The teaching information processing method provided in this disclosure first receives user input information providing feedback on the current teaching content; second, it performs semantic and intent recognition on the user input information to obtain the user's cognitive state regarding relevant nodes in the teaching knowledge graph; the teaching knowledge graph is a graph representing the relationship between teaching concept nodes and teaching process nodes; third, based on the cognitive state and user input information, it updates the historical dialogue information and mastery information of target concept nodes recorded in the user's dialogue profile, and obtains the current focus node from the user dialogue profile; fourth, based on the mastery information, it selects a target node related to the current focus node in the teaching knowledge graph, and uses the target node as the current focus node in the user dialogue profile; then, based on the target node and the teaching knowledge graph, it generates a content constraint set for content generation; finally, based on the content constraint set and historical dialogue information, it generates the next teaching content information, uses the next teaching content information as the current teaching content information, and sends the current teaching content information to the user. This achieves a personalized adaptive teaching closed loop based on the knowledge graph, continuously pushing content that matches the user's mastery information and focus, significantly improving the relevance of teaching and learning efficiency.

[0050] In some optional implementations of this disclosure, the above-mentioned selection of a target node related to the current focus node in the teaching knowledge graph based on mastery information, and the selection of the target node as the current focus node in the user dialogue profile, includes: in response to a mastery level value in the mastery information being greater than a first threshold, selecting a target node from the successor nodes starting from the current focus node in the teaching knowledge graph based on node priority and the mastery level value in the mastery information; in response to a mastery level value in the mastery information being less than a second threshold, selecting a target node from the teaching knowledge graph associated with the current focus node and user-inputted erroneous cognitive nodes, and if no matching erroneous cognitive node is found, determining the target node based on the preceding dependent nodes of the current focus node; in response to a mastery level value in the mastery information being greater than or equal to the second threshold and less than or equal to the first threshold, selecting an example or analogy node associated with the current focus node and not yet used from the teaching knowledge graph as the target node, and if no example or analogy node is found, using the current focus node as the target node; and using the target node as the current focus node in the user dialogue profile, wherein the first threshold is greater than the second threshold.

[0051] In this optional implementation, the first threshold and the second threshold can be set based on development requirements, such as the first threshold being 0.9 and the second threshold being 0.4.

[0052] In this optional implementation, the selection of the target node from the successor nodes starting from the current focus node in the teaching knowledge graph, based on node priority and the mastery level value in the mastery information, includes: selecting nodes in the teaching knowledge graph that have a directional relationship with the current focus node as successor nodes, and selecting the node with the highest priority and the lowest mastery level value in the mastery information as the target node. Alternatively, it can be implemented by scoring based on priority and the mastery level value in the mastery information, and selecting the target node based on the score.

[0053] In this optional implementation, the error perception node is a teaching process node, belonging to the Misconception type in the teaching knowledge graph. The prerequisite dependency node is a teaching process node, belonging to the Prerequisite type in the teaching knowledge graph. The example or category node is a teaching process node, belonging to the Analogy type in the teaching knowledge graph.

[0054] Optionally, compare the mastery degree value in the mastery information with two preset threshold values T1 (the first threshold) and T2 (the second threshold, T1 > T2): When the mastery degree value in the mastery information > T1, retrieve the successor node set of this node in the teaching knowledge graph, and construct a score by integrating the teaching priority of the node and the mastery degree value in the user's mastery information of this node (such as priority weighting, adding points to nodes with low mastery degree values in the mastery information), and select the one with the highest score as the target node; when the mastery degree value in the mastery information < T2, analyze the user's recent input, and preferentially select the target node from the misperception nodes associated with the current node according to the association strength and priority; if there is no matching misconcept, then roll back to the previous (prerequisite) node, and determine the target node according to the dependency level and priority; when T2 ≤ the mastery degree value in the mastery information ≤ T1, select the target node from the example / analogy nodes associated with the current node and not yet used according to the coverage or novelty. If there are no available examples / analogies, then use the current node as the target node; finally, replace the current focused node in the user dialogue portrait with the target node, and synchronously update the usage mark, learning path and the record of the mastery degree value in the node mastery information in the portrait to provide context for subsequent teaching rounds.

[0055] Optionally, the above-mentioned selection of the target node related to the current focused node in the teaching knowledge graph based on the mastery information and using the target node as the current focused node in the user dialogue portrait includes: in response to detecting that the user has a high mastery degree of the current focused node based on the mastery degree value in the mastery information, select a teaching concept node related to the current focused node from the teaching knowledge graph, use the teaching concept node as the target node, and use the target node as the current focused node in the user dialogue portrait; in response to detecting that the user has a low mastery degree of the current focused node based on the mastery degree value in the mastery information, select a teaching process node related to the current focused node from the teaching knowledge graph, use the teaching process node as the target node and use the target node as the current focused node in the user dialogue portrait.

[0056] In some optional implementation manners of the present disclosure, the above-mentioned constraint set at least includes encouraging words, prohibited words and required words. The generation of the content constraint set for content generation based on the target node and the teaching knowledge graph includes: in response to the node type of the target node not being a question node or a misperception node, extract keywords from the metadata of the target node recorded in the teaching knowledge graph as encouraging words; in response to the target node being a question node, obtain the corresponding answer keywords from the teaching knowledge graph and use them as prohibited words; in response to the target node being a misperception node, extract error-related keywords from the metadata of the target node recorded in the teaching knowledge graph, and use the error-related keywords as required words.

[0057] In this optional implementation, the problem node is a teaching process node, the misconception node is a teaching concept node, the problem node is a node of type Question in Table 1, and the misconception node is a node of type Misconception in Table 1.

[0058] In this optional implementation, the system takes the target node and the teaching knowledge graph as input. First, it reads the type of the target node and initializes the content constraint set (encouraged words, prohibited words, and required words). When the target node is neither a problem node nor a misconception node, the system extracts the core subject words from the node's metadata (such as title, abstract, tags, prerequisite / related concepts) in the graph through keyword extraction and entity recognition (such as TF-IDF / RAKE / NER). After cleaning and removing stop words and noise, the encouraged words are written into the graph. When the target node is a problem node, the system extracts the answer points based on the problem-answer association or the "answer keyword" field in the graph. After synonym normalization and word form restoration, prohibited words are added to avoid revealing the answer. When the target node is a misconception node, the system extracts error-related keywords from its metadata and the "common misunderstandings / error descriptions" attribute. After normalization, these keywords are used as required words to ensure the correctness of the error correction. Then, the system performs deduplication and conflict resolution on the three types of words (such as required words taking precedence over prohibited words, and conflict items being pruned according to the node type strategy). The system also sets the upper limit for the number and length of these words and finally outputs the constraint set used for content generation.

[0059] Optionally, the above-mentioned generation of a content constraint set for content generation based on the target node and the teaching knowledge graph may further include: in response to the target node being a correct concept node, extracting keywords from the metadata of the correct concept node from the teaching knowledge graph as encouraged words in the content constraint set; in response to the target node being a teaching process node, detecting that the teaching process node is a problem node or a misconception node, and using keywords from the metadata of the problem node or misconception node as prohibited words in the content constraint set.

[0060] In some optional implementations of this disclosure, the above-mentioned generation of next teaching content information based on the content constraint set and historical dialogue information, the use of the next teaching content information as the current teaching content information, and the sending of the current teaching content information to the user includes: encoding historical dialogue information into an input tag sequence; initializing the generated tag sequence to be empty; establishing a to-do set based on the required words in the constraint set; and cyclically executing the following steps, provided that the generation step does not exceed the maximum length, where the maximum length is the maximum number of lexical units: concatenating the input tag sequence with the currently generated tag sequence to form a context input large language model, and obtaining the candidate category score vector of the large language model in the generation step, wherein the dimension of the candidate category score vector is the same as the size of the vocabulary. Small; Based on the content constraint set, generate a constraint bias vector with the same dimension as the candidate category score vector; Based on the candidate category score vector and the constraint bias vector, calculate the corrected score vector; Based on the corrected score vector, obtain the probability distribution; Based on the probability distribution, select the next word and put the next word into the generated tag sequence; In response to the next word belonging to the to-do set, remove the next word from the to-do set; If the to-do set is not cleared, prohibit the end symbol from being used as the next word; If the generation step is not equal to the preset maximum length, decode the generated tag sequence to obtain the next teaching content information, use the next teaching content information as the current teaching content information, and send the current teaching content information to the user.

[0061] In this optional implementation, the vocabulary dimension is the same as the dimension of the large language model. During the autoregressive generation process of the large language model, at each step... A vocabulary size will be calculated ( The candidate class score (logits) vector The candidate category score is the "raw output" of the last layer of the large language model. The probability distribution of the standard next token is shown in Equation (1):

[0062] (1)

[0063] In this optional implementation, a constraint bias vector with the same dimension as the candidate category score is introduced through the content constraint set. The candidate category scores are adjusted to obtain the adjusted score vector. As shown in equation (2):

[0064] (2)

[0065] Therefore, the probability distribution of the final tokens is derived as shown in equation (3):

[0066] (3)

[0067] As shown in Figure 2, the knowledge graph constraint generation engine, as the core technology of the system, is responsible for applying the structured constraints of the teaching knowledge graph to the large language model generation process in real time, so as to achieve precise control over the output of the large language model and ensure the scientific nature of the teaching content.

[0068] In this optional implementation, the prompt words are first encoded into an input token sequence and the generated token sequence is set to empty. The to-do set is initialized based on the required words in the constraint set. Then, under the condition that the number of generation steps does not exceed the preset maximum number of lexical units, the following loop is used: the input sequence and the generated sequence are concatenated to form a large language model context, the original candidate scores with the same dimension as the vocabulary are calculated, the constraint bias vector with the same dimension is generated based on the content constraint set, the candidate scores are weighted and normalized to obtain a probability distribution, the next lexical unit is selected from it and appended to the generated sequence; if the lexical unit belongs to the to-do set, it is removed, and the selection of the end symbol is prohibited when the to-do set is not cleared to ensure that the constraints are satisfied; when the stopping condition is met (such as the to-do set is cleared or other set conditions are met) and the maximum length is not reached, the generated sequence is decoded to obtain the next teaching content information, which is then used as the new current teaching content information and sent to the user.

[0069] In some optional implementations of this disclosure, each dimension of the aforementioned constraint bias vector corresponds one-to-one with each lexical unit in the vocabulary. The aforementioned generation of constraint bias vectors with the same dimension as the candidate category score vector based on the content constraint set includes: in response to a lexical unit belonging to an encouraged word in the content constraint set, the constraint bias vector corresponding to that word is a positive incentive weight; in response to a lexical unit belonging to a prohibited word in the content constraint set, the constraint bias vector corresponding to that word is negative infinity; in response to a lexical unit belonging to a required word in the content constraint set, the constraint bias vector corresponding to that word is a strong guiding weight; in response to a lexical unit not belonging to any of the encouraged words, prohibited words, or required words in the content constraint set, the constraint bias vector corresponding to that word is zero.

[0070] In this optional implementation, the constraint bias vector Construction method:

[0071] vector Each dimension A token corresponding to the vocabulary list Its value is set according to the content constraint set, as shown in equation (4):

[0072] (4)

[0073] In equation (4), This represents the positive incentive weight (e.g., alpha = 5.0), which can significantly increase the probability of encouraging words appearing. This indicates absolute prohibition, which is negative infinity. In engineering implementation, this is a very large negative number, and negative infinity makes the softmax probability of this word close to 0. This indicates a strong guiding weight (e.g., beta = 10.0) used to ensure that key information must appear in the generated text. This weight can be dynamically set to 0 after the relevant words are generated.

[0074] In this optional implementation, at each generation step of decoding, the system constructs a constraint bias vector with the same dimension as the candidate category score vector for the vocabulary based on a predefined set of content constraints: the constraint bias vector is initialized to all zeros; positive incentive weights are assigned to lexical units belonging to "encouraged words"; stronger guiding weights are assigned to lexical units belonging to "required words"; negative infinity is assigned to lexical units belonging to "prohibited words," and the remaining lexical units remain at 0. Subsequently, the constraint bias vector is added element-wise to the candidate category score vector output by the model to obtain a corrected score vector, which is then subjected to softmax to obtain the constrained probability distribution for sampling or greedy selection; the positive incentive weights and strong guiding weights can be set as hyperparameters or learned through data-driven methods, and can be dynamically updated according to the constraint satisfaction state during the generation process (e.g., when a required word has been generated, its corresponding weight is reset to zero), so as to enhance the probability of the occurrence of encouraged and required items while ensuring that prohibited items are not selected.

[0075] In some optional implementations of this disclosure, the aforementioned cognitive state includes: node names, level of understanding, or correctness of answers in the knowledge graph; updating historical dialogue information and mastery information of target concept nodes recorded in the user's user dialogue profile based on the cognitive state and user input information; and obtaining the current focus node from the user dialogue profile includes: determining and updating the mastery information of target concept nodes in the user's user dialogue profile based on node names, level of understanding, or correctness of answers in the knowledge graph; updating historical dialogue information in the user dialogue profile based on user input information; and obtaining the current focus node from the user dialogue profile.

[0076] In this optional implementation, the system first performs entity / relationship linking on the user input, mapping the input to one or more nodes in the knowledge graph, and obtains the cognitive state {node name N}. i Confidence level c i Comprehension level / correctness of answer score s i ∈[0,1]}; then, the user's dialogue profile is loaded using the user ID as the key, and a record is appended to the historical dialogue information (timestamp t, original input, matching node set, system response, s). i c i ), and update the information according to nodes: for each N i Using exponential sliding or Bayesian updates, for example, m i ←(1−n)·mi +n·s i (n defaults to 0.3), synchronously maintain the occurrence count and the most recent interaction time t. i To obtain the current focus node, calculate the comprehensive focus score f of the candidate nodes. i =λ1·c i +λ2·r i +λ3·(1−m i ), where r i =exp(−Δt i / τ) is the time-sensitivity weight (τ defaults to 30 minutes), λ1+λ2+λ3=1, and f is selected. i The largest node is the current focus node; when the candidate f value is close to (|Δf|<δ, δ defaults to 0.05), the decision is made based on the frequency of occurrence in the most recent dialogue window or the latest time; finally, the updated mastery information and historical records are persisted and written back to the user dialogue profile to drive subsequent personalized question answering and recommendation.

[0077] Based on the above embodiments, the implementation process of the teaching information processing method will be described below with reference to a specific example in Figure 2:

[0078] User A's user profile includes the following information regarding their mastery of target concept nodes: concept_current: { "mastery": 0.95, "status": "mastered"}, concept_voltage: { "mastery": 0.92, "status": "mastered"}, concept_resistance: { "mastery": 0.91, "status": "mastered"}, concept_ohms_law: { "mastery": 0.1, "status": "unknown"}.

[0079] Current teaching content information: "We have already learned about current, voltage, and resistance. Do you think there might be some hidden relationship between these three friends? Feel free to guess!"; User A's input: "I guess the higher the voltage, the higher the resistance." The dialogue strategy manager recognizes this as a common misconception. Update user A's user dialogue profile, reduce the mastery value in the concept_resistance mastery information, and mark the status in the mastery information as confused; The dialogue strategy manager queries the teaching knowledge graph and finds a misconception node misconception_voltage_affects_resistance related to "voltage affects resistance". The dialogue strategy manager determines the target node as misconception_voltage_affects_resistance and generates a constraint set constraints = {"encourage": ["conductor", "property", "material"], "forbid":["Ohm's Law"], "must_include": ["voltage", "resistance"]} for the knowledge graph constraint generation engine. The purpose of this set of constraints is to guide the conversation around the premise that "resistance is an inherent property of a conductor" and to require reference to the user's mention of "voltage" and "resistance" in order to formulate a targeted response.

[0080] The knowledge graph constraint generation engine receives historical dialogue information and constraints. During decoding, the logits of words such as "voltage" and "resistance" are assigned a strong guiding weight beta=10.0 to ensure their appearance in sentences. The engine outputs the following to the dialogue interface: "This is a fantastic conjecture! Many students would think this way at first. However, let's consider this: the resistance of a conductor is more like its own 'personality,' such as its height and weight, usually determined by its material and shape. Do you think that if the voltage applied across it changes, its 'personality' will change accordingly?" This serves as the current teaching content.

[0081] Effect Analysis: This passage, through analogy, gently refutes the user's erroneous assumptions while precisely guiding them to think about the intrinsic properties of resistance, which is entirely in line with the principles of Socratic teaching.

[0082] Subsequently, with the user's correct answers, the system will gradually introduce the "direct proportion" relationship, then the inverse proportion of current to resistance when resistance remains constant, ultimately guiding the user to summarize the complete law that current is directly proportional to voltage and inversely proportional to resistance. Only after confirming that user A fully understands will the dialogue policy manager update the status of concept_ohms_law in the user's dialogue profile to mastered, and formally provide "Ohm's Law" and the formula I=U / R as a summary.

[0083] As can be seen from the above process, the system corresponding to the teaching information processing method disclosed herein is not a simple question-and-answer robot. It is a dynamic and strategic teaching guidance system. Each step of the dialogue is carefully decided by the dialogue strategy manager based on the user's cognitive state and the teaching knowledge graph. Then, through the core innovative module of the knowledge graph-constrained generation engine, the teaching intention is accurately transformed into natural, fluent, and inspiring language.

[0084] In some optional implementations of this disclosure, the method further includes: responding to the current teaching content information being empty and the user logging into the system for the first time, determining the teaching start node based on the preset learning outline and the user's mastery information of the target concept nodes in the user's dialogue profile; based on the mastery information of the teaching start node, selecting graph nodes related to the teaching start node in the teaching knowledge graph using a node teaching strategy, and using the graph nodes as the current focus nodes; generating a content constraint set for content generation based on the graph nodes and the teaching knowledge graph; generating the current teaching content information based on the graph nodes and the content constraint set, and sending the current teaching content information to the user.

[0085] In this optional implementation, when the current teaching content information is detected to be empty and the user is logging in for the first time, the starting node determination module is invoked. Based on the knowledge point dependencies represented by a directed acyclic graph in the preset learning syllabus and the mastery information of each node in the user's dialogue profile, the starting node is selected from the candidates with the lowest scores, provided that the prerequisite dependencies have been mastered. Subsequently, the node teaching strategy module selects adjacent prerequisite / same-level / advanced nodes (association degree ≥ α, difficulty matching) in the teaching knowledge graph according to the thresholds θ1 and θ2 of the mastery information of the starting node, and sets the selected node as the current focus node. The content constraint generation module combines the metadata of the focus node (learning objectives, keywords, prerequisite list, case library, difficulty tags) with the knowledge graph topology and user profile preferences (learning style, duration, language proficiency) to generate a set of content constraints, clarifying the objectives, scope, difficulty, knowledge boundaries, glossary, interaction format, assessment method, and length. Based on this, the content generation module uses a hybrid strategy of template splicing and natural language generation to generate the current teaching content information (including explanations, examples, exercises, and instant quizzes), and sends it to the user through a message push service, while also writing the session status and learning progress to support subsequent iterative teaching.

[0086] The following detailed explanation of the above embodiment is illustrated with a specific example in Figure 2: User Wes logs into the system corresponding to the teaching information processing method. In the user interaction layer of the system, the user accesses the dialogue interface. The system loads or initializes the user's dialogue profile through the user's identifier. The initial user dialogue profile shows that Wes has mastered basic concepts such as "current," "voltage," and "resistance" (mastery > 0.9), but is in an unknown state regarding "Ohm's Law" (mastery = 0.1, status = "unknown"). Based on the preset learning outline and the mastery information of Ohm's Law in the user's dialogue profile, the system determines that the teaching starting node for this dialogue is a teaching concept node of "Ohm's Law." The system queries the teaching knowledge graph and finds that this teaching concept node is associated with a heuristic question node q_initial_relation through a hasQuestion type edge. Therefore, q_initial_relation is set as the target node. Based on the target node, the current focus node in the user's dialogue profile is updated. Based on the mastery information of the target node, a content constraint set is generated: constraints = {"encourage": ["relation", "feel", "connection"], "forbid": ["Ohm's Law", ...} The constraint set, defined by "I=U / R", "directly proportional", and "inversely proportional", aims to encourage user thinking but prohibits the system from directly providing any form of answer or terminology. The knowledge graph constraint generation engine receives the prompt = "We have learned about current, voltage, and resistance." and the aforementioned constraint set sent by the dialogue policy manager. This controls the large language model during the generation process, with the logits vector being corrected in real-time: for example, the logits value of tokens containing words like "relationship" or "feel" is increased by alpha = 5.0; the logits value of tokens containing words like "Ohm's Law" or "directly proportional" is set to -inf. Finally, the knowledge graph constraint generation engine sends the current teaching content information to the user interaction layer. The user interaction layer outputs the current teaching content information to the user through the dialogue interface: "We have learned about current, voltage, and resistance. Do you think there might be some hidden relationship between these three friends? Feel free to guess!"

[0087] Referring further to Figure 3, as an implementation of the methods shown in the above figures, this disclosure provides an embodiment of a teaching information processing device, which corresponds to the method embodiment shown in Figure 1, and the device can be specifically applied to various electronic devices.

[0088] As shown in Figure 3, the teaching information processing device 300 provided in this embodiment includes: a receiving unit 301, an identification unit 302, an updating unit 303, a selection unit 304, a generation unit 305, and a sending unit 306. The receiving unit 301 can be configured to receive user input information providing feedback on the current teaching content. The identification unit 302 can be configured to perform semantic and intent recognition on the user input information to obtain the user's cognitive state regarding relevant nodes in the teaching knowledge graph; the teaching knowledge graph is a graph representing the relationship between teaching concept nodes and teaching process nodes. The updating unit 303 can be configured to update the historical dialogue information and mastery information of target concept nodes recorded in the user's user dialogue profile based on the cognitive state and user input information, and obtain the current focus node from the user dialogue profile. The selection unit 304 can be configured to select a target node related to the current focus node in the teaching knowledge graph based on the mastery information, and use the target node as the current focus node in the user dialogue profile. The aforementioned generation unit 305 can be configured to generate a set of content constraints for content generation based on the target node and the teaching knowledge graph. The aforementioned sending unit 306 can be configured to generate next teaching content information based on the target node and the content constraint set, use the next teaching content information as the current teaching content information, and send the current teaching content information to the user.

[0089] In this embodiment, the specific processing of the receiving unit 301, the identification unit 302, the updating unit 303, the selection unit 304, the generation unit 305, and the sending unit 306 in the teaching information processing device 300, and the resulting technical effects, can be referred to the relevant descriptions of steps 101, 102, 103, 104, 105, and 106 in the corresponding embodiment of Figure 1, which will not be repeated here.

[0090] In some embodiments of this disclosure, the selection unit 304 is configured to: in response to a mastery level value in the mastery information being greater than a first threshold, select a target node from the successor nodes starting from the current focus node in the teaching knowledge graph based on node priority and the mastery level value in the mastery information; in response to a mastery level value in the mastery information being less than a second threshold, select a target node from the teaching knowledge graph associated with the current focus node and user-inputted erroneous cognitive nodes; if no matching erroneous cognitive node is found, determine the target node based on the predecessor dependent nodes of the current focus node; in response to a mastery level value in the mastery information being greater than or equal to the second threshold and less than or equal to the first threshold, select an example or analogy node associated with the current focus node and not yet used from the teaching knowledge graph as the target node; if no example or analogy node is found, use the current focus node as the target node; and use the target node as the current focus node in the user dialogue profile, where the first threshold is greater than the second threshold.

[0091] In some embodiments of this disclosure, the constraint set includes at least encouraged words, prohibited words, and required words. The generation unit 305 is configured to: in response to the target node not being a problem node or a misconception node, extract keywords from the metadata of the target node recorded in the teaching knowledge graph as encouraged words; in response to the target node being a problem node, obtain the corresponding answer keywords from the teaching knowledge graph and use them as prohibited words; in response to the target node being a misconception node, extract error-related keywords from the metadata of the target node recorded in the teaching knowledge graph and use the error-related keywords as required words.

[0092] In some embodiments of this disclosure, the sending unit 306 is configured to: encode historical dialogue information into an input token sequence; initialize the generated token sequence to be empty; establish a to-do set based on the required words in the constraint set; and repeatedly execute the following steps, where the maximum length is the maximum number of lexical units, provided that the generation step does not exceed the maximum length: concatenate the input token sequence with the currently generated token sequence to form a context input large language model, and obtain the candidate category score vector of the large language model in the generation step, the dimension of the candidate category score vector and the vocabulary size; and generate constraints with the same dimension as the candidate category score vector based on the content constraint set. Bias vector; Calculate the corrected score vector based on the candidate category score vector and the constraint bias vector; Obtain the probability distribution based on the corrected score vector; Select the next word based on the probability distribution and add the next word to the generated tag sequence; If the next word belongs to the to-do set, remove the next word from the to-do set. If the to-do set is not cleared, the end symbol is prohibited from being used as the next word; If the generation step is not equal to the preset maximum length, decode the generated tag sequence to obtain the next teaching content information, use the next teaching content information as the current teaching content information, and send the current teaching content information to the user.

[0093] In some embodiments of this disclosure, each dimension of the constraint bias vector corresponds one-to-one with each lexical unit in the vocabulary. The sending unit 306 is further configured to: respond to a lexical unit belonging to an encouraged word in the content constraint set, the constraint bias vector corresponding to that word is a positive incentive weight; respond to a lexical unit belonging to a prohibited word in the content constraint set, the constraint bias vector corresponding to that word is negative infinity; respond to a lexical unit belonging to a required word in the content constraint set, the constraint bias vector corresponding to that word is a strong guiding weight; respond to a lexical unit not belonging to any of the encouraged words, prohibited words, or required words in the content constraint set, the constraint bias vector corresponding to that word is zero.

[0094] In some embodiments of this disclosure, the aforementioned cognitive state includes: node names in the knowledge graph, level of understanding, or correctness of the answer. The aforementioned update unit 303 is configured to: determine and update the mastery information of the target concept node in the user's user dialogue profile based on the node names in the knowledge graph, level of understanding, or correctness of the answer; update the historical dialogue information in the user dialogue profile based on user input information; and obtain the currently focused node from the user dialogue profile.

[0095] In some embodiments of this disclosure, the device 300 further includes an initial unit (not shown in the figure), which is configured to: in response to the current teaching content information being empty and the user logging into the system for the first time, determine a teaching start node based on a preset learning outline and the mastery information of target concept nodes in the user's dialogue profile; based on the mastery information of the teaching start node, select graph nodes related to the teaching start node in the teaching knowledge graph using a node teaching strategy, and use the graph nodes as the current focus nodes; generate a content constraint set for content generation based on the graph nodes and the teaching knowledge graph; generate current teaching content information based on the graph nodes and the content constraint set, and send the current teaching content information to the user.

[0096] The teaching information processing apparatus provided in the embodiments of this disclosure firstly receives user input information from the user regarding the current teaching content information; secondly, the identification unit 302 performs semantic and intent recognition on the user input information to obtain the user's cognitive state regarding relevant nodes in the teaching knowledge graph; the teaching knowledge graph is a graph representing the relationship between teaching concept nodes and teaching process nodes; thirdly, the updating unit 303 updates the historical dialogue information and mastery information of target concept nodes recorded in the user's user dialogue profile based on the cognitive state and user input information, and obtains the current focus node from the user dialogue profile; next, the selection unit 304 selects a target node related to the current focus node in the teaching knowledge graph based on the mastery information, and uses the target node as the current focus node in the user dialogue profile; then, the generation unit 305 generates a content constraint set for content generation based on the target node and the teaching knowledge graph; finally, the sending unit 306 generates the next teaching content information based on the target node and the content constraint set, uses the next teaching content information as the current teaching content information, and sends the current teaching content information to the user. This enables a personalized, adaptive teaching loop centered around a knowledge graph, dynamically adjusting focus nodes and content constraints based on the user's information, and continuously pushing content that better suits the learning state, thereby improving the relevance, coherence, and efficiency of teaching.

[0097] The collection, storage, use, processing, transmission, provision, and disclosure of user personal information involved in the technical solution disclosed herein comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0098] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0099] Figure 4 illustrates a schematic block diagram of an example electronic device 400 that can be used to implement embodiments of the present disclosure. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their patterns are merely examples and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0100] As shown in Figure 4, device 400 includes a computing unit 401, which can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) 402 or a computer program loaded from storage unit 408 into random access memory (RAM) 403. RAM 403 may also store various programs and data required for the operation of device 400. The computing unit 401, ROM 402, and RAM 403 are interconnected via bus 404. Input / output (I / O) interface 405 is also connected to bus 404.

[0101] Multiple components in device 400 are connected to I / O interface 405, including: input unit 406, such as keyboard, mouse, etc.; output unit 407, such as various types of monitors, speakers, etc.; storage unit 408, such as disk, optical disk, etc.; and communication unit 409, such as network card, modem, wireless transceiver, etc. Communication unit 409 allows device 400 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0102] The computing unit 401 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 401 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 401 performs the various methods and processes described above, such as instructional information processing methods. For example, in some embodiments, the instructional information processing method may be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 408. In some embodiments, part or all of the computer program may be loaded and / or installed on device 400 via ROM 402 and / or communication unit 409. When the computer program is loaded into RAM 403 and executed by the computing unit 401, one or more steps of the instructional information processing method described above may be performed. Alternatively, in other embodiments, the computing unit 401 may be configured to perform instructional information processing methods by any other suitable means (e.g., by means of firmware).

[0103] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0104] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to the processor or controller of a general-purpose computer, special-purpose computer, or other programmable instructional information processing device, such that when executed by the processor or controller, the patterns / operations specified in the flowcharts and / or block diagrams are implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0105] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0106] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0107] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0108] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0109] The foregoing description of specific exemplary embodiments of this disclosure is for illustrative and explanatory purposes. These descriptions are not intended to limit this disclosure to the precise forms disclosed, and it will be apparent that many changes and variations can be made in accordance with the foregoing teachings. The exemplary embodiments were chosen and described in order to explain the specific principles of this disclosure and their practical application, thereby enabling those skilled in the art to implement and utilize various different exemplary embodiments of this disclosure, as well as various different choices and variations. The scope of this disclosure is intended to be defined by the claims and their equivalents.

Claims

1. A method for processing teaching information, the method comprising: Receive user input information to provide feedback on the current teaching content; The user input information is semantically and intentally recognized to obtain the user's cognitive state regarding relevant nodes in the teaching knowledge graph. The teaching knowledge graph is a graph representing the relationship between teaching concept nodes and teaching process nodes. Based on the cognitive state and the user input information, the historical dialogue information and mastery information of target concept nodes recorded in the user's user dialogue profile are updated, and the current focus node is obtained from the user dialogue profile. The target concept node is a teaching concept node that needs to be learned and is pre-loaded according to the user's learning syllabus in the teaching knowledge graph, and the current focus node is a node in the teaching knowledge graph that the teaching system focuses on in the current dialogue round. Based on the acquired information, a target node related to the current focus node is selected in the teaching knowledge graph, and the target node is used as the current focus node in the user dialogue profile. Based on the target node and the teaching knowledge graph, generating a content constraint set for content generation includes: in response to the target node not being a problem node or a misconception node, extracting keywords from the metadata of the target node recorded in the teaching knowledge graph as encouraged words; in response to the target node being a problem node, obtaining corresponding answer keywords from the teaching knowledge graph and using them as prohibited words; in response to the target node being a misconception node, extracting error-related keywords from the metadata of the target node recorded in the teaching knowledge graph and using these error-related keywords as required words; the constraint set includes at least encouraged words, prohibited words, and required words; based on the content constraint set and the historical dialogue information, generating next teaching content information, using the next teaching content information as the current teaching content information, and sending the current teaching content information to the user includes: encoding the historical dialogue information into an input tag sequence; initializing the generated tag sequence to be empty; establishing a to-do set based on the required words in the constraint set; and ensuring that the generation step does not exceed the maximum length. Under the condition of [condition], the following steps are executed cyclically, where the maximum length is the maximum number of lexical units: The input tag sequence is concatenated with the currently generated tag sequence to form a context input large language model, and the candidate category score vector of the large language model in the generation step is obtained, the dimension of the candidate category score vector being the same as the vocabulary size; Based on the content constraint set, a constraint bias vector with the same dimension as the candidate category score vector is generated; Based on the candidate category score vector and the constraint bias vector, a corrected score vector is calculated; Based on the corrected score vector, a probability distribution is obtained; Based on the probability distribution, the next lexical unit is selected and placed into the generated tag sequence; In response to the next lexical unit belonging to the to-do set, the next lexical unit is removed from the to-do set, and if the to-do set is not cleared, the end character is prohibited from being used as the next lexical unit; If the generation step is not equal to the preset maximum length, the generated tag sequence is decoded to obtain the next teaching content information, the next teaching content information is used as the current teaching content information, and the current teaching content information is sent to the user.

2. The method according to claim 1, wherein, The step of selecting a target node related to the current focus node in the teaching knowledge graph based on the mastery information, and using the target node as the current focus node in the user dialogue profile, includes: responding to a mastery level value in the mastery information being greater than a first threshold, selecting a target node from the successor nodes starting from the current focus node in the teaching knowledge graph based on node priority and the mastery level value in the mastery information; responding to a mastery level value in the mastery information being less than a second threshold, selecting the target node from the teaching knowledge graph associated with the current focus node and the user-inputted erroneous cognitive nodes; if no matching erroneous cognitive node is found, determining the target node based on the current focus node's preceding dependent nodes; responding to a mastery level value in the mastery information being greater than or equal to the second threshold and less than or equal to the first threshold, selecting an example or analogy node associated with the current focus node and not yet used from the teaching knowledge graph as the target node; if no example or analogy node is found, using the current focus node as the target node; and using the target node as the current focus node in the user dialogue profile, wherein the first threshold is greater than the second threshold.

3. The method according to claim 1, wherein, Each dimension of the constraint bias vector corresponds one-to-one with each lexical unit in the vocabulary. The step of generating a constraint bias vector with the same dimension as the candidate category score vector based on the content constraint set includes: in response to a lexical unit belonging to an encouraged word in the content constraint set, the constraint bias vector corresponding to that word is a positive incentive weight; in response to a lexical unit belonging to a prohibited word in the content constraint set, the constraint bias vector corresponding to that word is negative infinity; in response to a lexical unit belonging to a required word in the content constraint set, the constraint bias vector corresponding to that word is a strong guiding weight; in response to a lexical unit not belonging to any of the encouraged, prohibited, or required words in the content constraint set, the constraint bias vector corresponding to that word is zero.

4. The method according to claim 1, wherein, The cognitive state includes: node names in the knowledge graph, level of understanding, or correctness of the answer. The step of updating the historical dialogue information and mastery information of the target concept node recorded in the user's user dialogue profile based on the cognitive state and the user input information, and obtaining the current focus node from the user dialogue profile, includes: determining and updating the mastery information of the target concept node in the user's user dialogue profile based on the node names, level of understanding, or correctness of the answer in the knowledge graph; updating the historical dialogue information in the user dialogue profile based on the user input information; and obtaining the current focus node from the user dialogue profile.

5. The method according to any one of claims 1-4, wherein, The method further includes: responding to the current teaching content information being empty and the user logging into the system for the first time, determining a teaching start node based on a preset learning outline and the mastery information of target concept nodes in the user's dialogue profile; based on the mastery information of the teaching start node, selecting graph nodes related to the teaching start node in the teaching knowledge graph using the teaching strategy of the teaching start node, and using the graph nodes as the current focus nodes; generating a content constraint set for content generation based on the graph nodes and the teaching knowledge graph; generating current teaching content information based on the graph nodes and the content constraint set, and sending the current teaching content information to the user.

6. A teaching information processing device, the device comprising: The receiving unit is configured to receive user input information that provides feedback on the current teaching content. The recognition unit is configured to perform semantic and intent recognition on the user input information to obtain the user's cognitive state regarding relevant nodes in the teaching knowledge graph; The instructional knowledge graph is a graph representing the relationship between instructional concept nodes and instructional process nodes; the update unit is configured to update the historical dialogue information and mastery information of the target concept node recorded in the user's user dialogue profile based on the cognitive state and the user input information, and obtain the current focus node from the user dialogue profile; the target concept node is the instructional concept node that needs to be learned, pre-loaded according to the user's learning syllabus in the instructional knowledge graph, and the current focus node is the node in the instructional knowledge graph that the instructional system focuses on in the current dialogue round; The selection unit is configured to select a target node related to the current focus node in the teaching knowledge graph based on the mastered information, and to use the target node as the current focus node in the user dialogue profile. The generation unit is configured to generate a set of content constraints for content generation based on the target node and the teaching knowledge graph; the constraint set includes at least encouraged words, prohibited words, and required words; the generation unit is further configured to extract keywords from the metadata of the target node recorded in the teaching knowledge graph as encouraged words in response to the target node not being a problem node or a misconception node; and to obtain corresponding answer keywords from the teaching knowledge graph as prohibited words in response to the target node being a problem node. In response to the target node being an erroneous cognitive node, error-related keywords are extracted from the metadata of the target node recorded in the teaching knowledge graph, and these error-related keywords are designated as mandatory words. The sending unit is configured to generate next teaching content information based on the target node and the content constraint set, use the next teaching content information as the current teaching content information, and send the current teaching content information to the user; the sending unit is further configured to: encode the historical dialogue information into an input tag sequence; initialize the generated tag sequence to be empty; A to-do set is established based on the required words in the constraint set; the following steps are executed iteratively without exceeding the maximum length in the generation step, where the maximum length is the maximum number of lexical units: the input tag sequence is concatenated with the currently generated tag sequence to form a context input large language model, and the candidate category score vector of the large language model in the generation step is obtained, wherein the dimension of the candidate category score vector is the same as the vocabulary size; a constraint bias vector with the same dimension as the candidate category score vector is generated based on the content constraint set; a corrected score vector is calculated based on the candidate category score vector and the constraint bias vector; and a probability distribution is obtained based on the corrected score vector. Based on the probability distribution, the next word is selected and placed into the generated token sequence; In response to the next word belonging to the to-do set, the next word is removed from the to-do set. When the to-do set is not cleared, the end symbol is prohibited from being the next word. When the generation step is not equal to the preset maximum length, the generated tag sequence is decoded to obtain the next teaching content information. The next teaching content information is used as the current teaching content information and sent to the user.

7. An electronic device, characterized in that, include: At least one processor; And a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-5.

8. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-5.

Images