Interactive teaching method, device, equipment, storage medium and program product

Abstract

The invention relates to the technical field of artificial intelligence, and provides an interactive teaching method and device, equipment, a storage medium and a program product. The method comprises the steps of creating a dynamic thinking map of a user based on a to-be-answered question; receiving an input text of a user; inputting the dynamic thinking map, the input text and the historical dialogue of the user into a large language model to obtain a real-time thinking map; selecting a target teaching strategy from a plurality of teaching strategies based on the real-time thinking map; inputting the target teaching strategy, the input text, the historical dialogue and the real-time thinking map into a large language model to obtain a teaching reply text; and taking the real-time thinking map as a dynamic thinking map, and returning to the step of receiving the input text of the user until the user successfully solves the question to be answered. Through the above mode, the method can comply with the actual thinking condition of the user in the problem solving process, dynamically provides reasonable teaching guidance for the user according to the real-time thinking map in the problem solving process of the user, and is beneficial to improving the teaching quality of interactive teaching.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to an interactive teaching method, device, equipment, storage medium, and program product. Background Technology

[0002] With the deep integration of information technology and education, computer-aided instruction systems have become an important tool for improving users' learning efficiency. Especially in subjects such as mathematics and physics that rely on rigorous logical reasoning and abstract thinking, developing intelligent teaching systems that can effectively guide users' thinking and cultivate their problem-solving abilities has always been the core research goal.

[0003] In recent years, the emergence of Large Language Models (LLMs) has brought about a paradigm shift in intelligent teaching systems. Interactive teaching systems based on LLMs have emerged, which can engage in fluent and open natural language dialogues with users, improving the flexibility and humanization of teaching interactions and demonstrating enormous application potential.

[0004] Currently, interactive teaching systems based on large language models mainly employ two teaching methods: those based on historical dialogues and those based on preset reference answers. In the historical dialogue-based method, the interactive teaching system generates teaching texts based on the user's historical dialogues. However, this method struggles to accurately analyze the user's thought process and cannot provide appropriate guidance based on that, resulting in poor teaching quality. In the preset reference answer-based method, the interactive teaching system generates teaching texts that guide the user towards the reference answer based on the gap between the user's provided answer and the preset reference answer. However, this method is closed, only able to teach according to the pre-set reference answer's line of thought. The teaching model is relatively fixed and cannot provide corresponding guidance based on the user's actual thinking, easily limiting the user's creative thinking and further leading to poor teaching quality.

[0005] In summary, existing interactive teaching methods fail to provide adequate instructional guidance, resulting in poor teaching quality. Summary of the Invention

[0006] This application provides an interactive teaching method, apparatus, device, storage medium, and program product to solve the technical problem that existing interactive teaching methods are unable to provide reasonable teaching guidance, resulting in poor teaching quality.

[0007] In a first aspect, embodiments of this application provide an interactive teaching method, comprising: creating a dynamic mind map for the user based on a question input by the user; the dynamic mind map representing the user's problem-solving logic path; receiving input text from the user; the input text describing the user's thought process in solving the question; inputting the dynamic mind map, the input text, and the user's historical dialogue into a large language model to update the mind map and obtain a real-time mind map output by the large language model; selecting a target teaching strategy from multiple teaching strategies based on the real-time mind map; inputting the target teaching strategy, the input text, the historical dialogue, and the real-time mind map into the large language model to obtain a teaching response text output by the large language model; using the real-time mind map as the dynamic mind map and returning to the step of receiving the user's input text, until the user successfully solves the question.

[0008] In one embodiment, a target teaching strategy is selected from multiple teaching strategies based on a real-time mind map, including: extracting features from the real-time mind map to generate teaching state features of the real-time mind map; calculating the posterior probability of each teaching strategy in the teaching strategy library under the teaching state features based on Bayes' theorem; and randomly sampling each teaching strategy in the teaching strategy library based on the posterior probability of each teaching strategy under the teaching state features to select the target teaching strategy.

[0009] In one embodiment, the real-time mind map includes at least two interactive nodes, and the teaching status features include frontier exploration width, node salience, and progress stagnation. The frontier exploration width is used to characterize the user's thinking tendency; the node salience is used to characterize the importance of the user's current interactive node; the current interactive node is the interactive node selected by the user in the real-time mind map; and the progress stagnation is used to characterize the fluency of the user's problem-solving process.

[0010] In one embodiment, the frontier exploration width is determined based on the number of active leaf nodes and the maximum depth of the real-time mind graph. The number of active leaf nodes is the number of recently created interactive nodes without child nodes in the real-time mind graph. Node salience is determined based on the in-degree and semantic weight of the current interactive node. Progress stagnation is determined based on the user's interaction rounds.

[0011] In one embodiment, the real-time mind graph is a directed acyclic graph representing the user's real-time problem-solving logic path; wherein, the real-time mind graph includes at least two interactive nodes and at least one edge, an edge is used to connect the two interactive nodes, an interactive node is used to represent a reasoning step in the user's problem-solving process, and an edge is used to represent the logical dependency relationship between the two interactive nodes.

[0012] In one embodiment, each interaction node includes strategy information, reasoning information, and result information; wherein, strategy information is used to describe the problem-solving strategy adopted by the corresponding reasoning step; reasoning information is used to describe the logical reasoning process of the corresponding reasoning step; and result information is used to describe the reasoning result of the corresponding reasoning step.

[0013] Secondly, embodiments of this application provide an interactive teaching device, comprising: an initialization module for creating a dynamic mind map of the user based on a question input by the user; the dynamic mind map represents the user's problem-solving logic path; a receiving module for receiving the user's input text; the input text is text describing the user's thought process in solving the question; a mind map updating module for inputting the dynamic mind map, the input text, and the user's historical dialogue into a large language model to update the mind map and obtain a real-time mind map output by the large language model; a teaching strategy selection module for selecting a target teaching strategy from multiple teaching strategies based on the real-time mind map; a generation module for inputting the target teaching strategy, the input text, the historical dialogue, and the real-time mind map into the large language model to obtain a teaching response text output by the large language model; and an interactive teaching module for using the real-time mind map as a dynamic mind map and returning to the step of receiving the user's input text until the user successfully solves the question.

[0014] Thirdly, embodiments of this application provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements any of the interactive teaching methods described above.

[0015] Fourthly, embodiments of this application provide a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements any of the interactive teaching methods described above.

[0016] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements any of the interactive teaching methods described above.

[0017] The interactive teaching method, apparatus, device, storage medium, and program product provided in this application introduce graph modeling technology into the user's problem-solving process. First, based on the user's input of the problem to be solved, a dynamic mind map representing the user's problem-solving logic path is created. Then, input text describing the user's thought process for solving the problem is received, and the dynamic mind map, input text, and the user's historical dialogue are input into a large language model. The large language model updates the dynamic mind map, generating a real-time mind map during the user's problem-solving process. Next, a target teaching strategy is selected from multiple teaching strategies based on the real-time mind map. The target teaching strategy, input text, historical dialogue, and real-time mind map are then input into the large language model, which generates a teaching response text. Finally, the real-time mind map is used as the dynamic mind map, and the process returns to the step of receiving the user's input text until the user successfully solves the problem. This interactive teaching method adapts to the user's actual thinking process during problem-solving, dynamically providing reasonable teaching guidance based on the user's real-time mind map, which is beneficial for improving the teaching quality of interactive teaching. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is one of the flowcharts of the interactive teaching method provided in the embodiments of this application.

[0020] Figure 2 This is the second flowchart of the interactive teaching method provided in the embodiments of this application.

[0021] Figure 3 This is a schematic diagram of the mind map update process provided in the embodiments of this application.

[0022] Figure 4 This is a schematic diagram illustrating the process of generating teaching response text provided in the embodiments of this application.

[0023] Figure 5 This is a schematic diagram of the algorithm flow for the mind map-based personalized interaction strategy selection algorithm provided in the embodiments of this application.

[0024] Figure 6 This is a schematic diagram of the mind map data model module provided in the embodiments of this application.

[0025] Figure 7 This is the third flowchart of the interactive teaching method provided in the embodiments of this application.

[0026] Figure 8 This is a schematic diagram of the structure of the interactive teaching device provided in the embodiments of this application.

[0027] Figure 9 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0029] Please see Figures 1 to 7 , Figure 1 This is one of the flowcharts illustrating the interactive teaching method provided in the embodiments of this application. Figure 2 This is the second flowchart illustrating the interactive teaching method provided in this application embodiment. Figure 3 This is a schematic diagram of the mind map updating process provided in an embodiment of this application. Figure 4 This is a schematic diagram illustrating the process of generating teaching response text provided in an embodiment of this application. Figure 5 This is a schematic diagram of the algorithm flow for the mind map-based personalized interaction strategy selection algorithm provided in the embodiments of this application. Figure 6 This is a schematic diagram of the mind map data model module provided in an embodiment of this application. Figure 7 This is the third flowchart of the interactive teaching method provided in the embodiments of this application.

[0030] In this embodiment of the application, the interactive teaching method is applied to an interactive teaching system based on a large language model and dynamic mind map (hereinafter referred to as the "system"). The system includes a mind map data model module, a mind map-based personalized interaction strategy selection algorithm module, and a context management module.

[0031] Among them, the mind map data model module is used to perform structured modeling of the user's non-linear problem-solving thinking process.

[0032] The personalized interaction strategy selection algorithm module is the decision-making center of the system. It is used to analyze the current state of the user's mind map and adaptively select the next teaching strategy accordingly.

[0033] The context management module is the core hub of the system, used to manage the interaction context between users and the system, ensuring that the large language model can accurately understand the dialogue context, update the mind map, and execute the selected teaching strategies.

[0034] like Figure 1 As shown, the interactive teaching method includes steps S110 to S160, and the specific steps are as follows: S110: Create a dynamic mind map for the user based on the unanswered question input by the user.

[0035] Dynamic mind maps are used to represent the user's problem-solving logic path.

[0036] Specifically, such as Figure 2 As shown, during the system startup and initialization phase, users (such as students) can input a question to be answered through the system's visual interface, thereby initiating a teaching interaction.

[0037] Optionally, the problem to be solved can be a mathematical problem, a physical problem, a chemical problem, etc.

[0038] Furthermore, in response to the user's input of an unanswered question, the system calls the application programming interface of the mind map data model module. The mind map data model module creates an initial dynamic mind map for the user based on the unanswered question, and uses the unanswered question itself as the root node (i.e., the initial interactive node) of the initial dynamic mind map.

[0039] Among them, dynamic mind maps are used to represent the user's problem-solving logic path.

[0040] Optionally, throughout the entire teaching interaction process, the system can continuously display the dynamic mind map to the user for reference or adjustment.

[0041] S120: Receive text input from the user.

[0042] The input text is a description of the user's thought process in solving the problem to be solved.

[0043] Specifically, after the system generates an initial dynamic mind map, users can begin to try to solve the problem by inputting text describing their thought process, ideas, or requests for solving the problem.

[0044] Optionally, users can select interactive nodes in the dynamic mind map through the system's visual interface to focus on the discussion content.

[0045] Optionally, the input text can be a description of a complete reasoning process, or a preliminary solution idea or direction that is not yet certain.

[0046] Optionally, when a user has no solution in mind, they can directly request the system to provide hints or guidance.

[0047] S130: Input the dynamic mind map, input text, and the user's historical dialogue into the large language model, update the mind map, and obtain the real-time mind map output by the large language model.

[0048] Specifically, such as Figure 3 As shown, the system's context management module responds to the user's input text by inputting the current dynamic mind map, the input text, and the user's historical dialogue into the large language model. The large language model can parse the user's input text and historical dialogue, and call the update interface functions (such as the create_node function and the create_edge function) defined by the mind map data model module to update the current dynamic mind map, generating and outputting the updated real-time mind map.

[0049] Among them, the real-time mind map is a directed acyclic graph that represents the user's real-time problem-solving logic path and can reflect the user's latest problem-solving ideas.

[0050] A real-time mind map consists of at least two interactive nodes and at least one edge. An edge connects the two interactive nodes. A node represents a reasoning step in the user's problem-solving process, and an edge represents the logical dependency between the two interactive nodes.

[0051] S140: Based on real-time mind maps, select the target teaching strategy from multiple teaching strategies.

[0052] Specifically, such as Figure 4 As shown, after generating the user's real-time mind map, the personalized interaction strategy selection algorithm module will run a personalized interaction strategy selection algorithm based on the mind map. This algorithm can analyze the mind map state of the real-time mind map, extract the teaching state characteristics of the real-time mind map, and select the most suitable target teaching strategy from multiple teaching strategies in the teaching strategy library based on the teaching state characteristics.

[0053] S150: Input the target teaching strategy, input text, historical dialogue and real-time mind map into the large language model to obtain the teaching response text output by the large language model.

[0054] Specifically, after determining the most suitable target teaching strategy, the context management module can use the target teaching strategy as a task instruction and input the target teaching strategy, input text, historical dialogue, and real-time mind map into the large language model. The large language model can then generate a guiding teaching response text based on the target teaching strategy, combined with the input text, historical dialogue, and real-time mind map.

[0055] S160: Convert the real-time mind map into a dynamic mind map and return to the step of receiving the user's input text until the user successfully solves the problem.

[0056] Specifically, after the large language model generates guided instructional response text, the instructional response text can be presented to the user, the real-time mind map can be used as a dynamic mind map, and the process can return to the step of receiving the user's input text, waiting for the user's next round of input, until the user successfully solves the question to be answered, and the complete dynamic mind map is also constructed.

[0057] The interactive teaching method provided in this application introduces graph modeling technology into the user's problem-solving process. First, based on the user's input of the problem to be solved, a dynamic mind map representing the user's logical path to solve the problem is created. Then, input text describing the user's thought process for solving the problem is received, and the dynamic mind map, input text, and the user's historical dialogue are input into a large language model. The large language model updates the dynamic mind map, generating a real-time mind map of the user's problem-solving process. Next, based on the real-time mind map, a target teaching strategy is selected from multiple teaching strategies. The target teaching strategy, input text, historical dialogue, and real-time mind map are then input into the large language model, which generates a teaching response text. Finally, the real-time mind map is used as the dynamic mind map, and the process returns to the step of receiving the user's input text until the user successfully solves the problem. This interactive teaching method adapts to the user's actual thinking process during problem-solving, providing dynamic and reasonable teaching guidance based on the user's real-time mind map, which is beneficial for improving the teaching quality of interactive teaching.

[0058] In some embodiments, selecting a target teaching strategy from multiple teaching strategies based on a real-time mind map includes: extracting features from the real-time mind map to generate teaching state features of the real-time mind map; calculating the posterior probability of each teaching strategy in the teaching strategy library under the teaching state features based on Bayes' theorem; and randomly sampling each teaching strategy in the teaching strategy library based on the posterior probability of each teaching strategy under the teaching state features to select the target teaching strategy.

[0059] After generating the user's real-time mind map, the system will run a personalized interaction strategy selection algorithm based on the mind map. This algorithm is a response strategy selection algorithm based on probability inference. It can perform real-time quantitative analysis on the real-time mind map and calculate the posterior probability of each teaching strategy in the teaching strategy library in the current situation, thereby realizing intelligent teaching strategy decision-making.

[0060] Specifically, such as Figure 5 As shown, before generating each guided instructional response text, the system can run a personalized interaction strategy selection algorithm based on mind maps to extract features from the real-time mind map and generate quantifiable teaching state features. .

[0061] Among them, teaching status characteristics It can reflect the user's current problem-solving progress and the user's thinking pattern.

[0062] Optionally, teaching status characteristics Including the breadth of frontier exploration Node saliency and progress stagnation .

[0063] Among them, the breadth of frontier exploration Used to characterize a user's thinking tendencies.

[0064] Node salience This is used to characterize the importance of the user's current interaction node, which is the interaction node selected by the user in the real-time mind map.

[0065] Stagnation of progress Used to characterize the fluency of the user's problem-solving process.

[0066] Furthermore, based on Bayes' theorem, characteristics of teaching states are established. With teaching strategies The probability mapping relationship between them is calculated for each teaching strategy in the teaching strategy library. Characteristics of teaching status Posterior probability .

[0067] Specifically, the system pre-sets a teaching strategy library that includes a variety of teaching strategies.

[0068] Optionally, the teaching strategies in the teaching strategy library include, but are not limited to, Socratic questioning strategy, tiered prompting strategy, role reversal strategy, open-ended selection strategy, and summarizing and clarifying strategy.

[0069] The Socratic questioning strategy refers to guiding users to examine the rigor of their conclusions and underlying assumptions by questioning their correct reasoning steps, thereby deepening their understanding of the knowledge points.

[0070] The hierarchical hint strategy refers to the system providing minimal help clues from macro-level ideas to specific steps when users get stuck in a problem-solving bottleneck, encouraging users to find a breakthrough on their own.

[0071] The role reversal strategy refers to inviting users to play the role of "teacher" and explain the reasoning process or principle of a certain step to the system in order to test whether users truly understand the relevant knowledge points.

[0072] An open-ended choice strategy refers to providing multiple parallel directions of thought or next steps at key crossroads in a user's problem-solving process, allowing the user to choose their own exploration path.

[0073] The summary and clarification strategy refers to summarizing and reviewing the completed problem-solving parts when the reasoning chain is long or the logic becomes complex, in order to consolidate the user's understanding and clarify potential ambiguities.

[0074] Furthermore, for each instructional strategy in the instructional strategy library According to Bayes' theorem, this teaching strategy Characteristics of teaching status Posterior probability It is proportional to the product of its likelihood and prior probability, i.e., it satisfies the following formula: ; in, Teaching strategies The prior probability, i.e., the teaching strategy The base probability used can be set based on general teaching experience; likelihood Indicating in teaching strategies Under the premise of optimal selection, the characteristics of the current teaching status were observed. The probability of.

[0075] Alternatively, the likelihood function can be set using a conditional probability table based on expert knowledge.

[0076] Specifically, for each teaching strategy in the teaching strategy library The system can provide this teaching strategy Predefine a conditional probability table based on expert knowledge to describe the probability of the teaching strategy. When applicable, characteristics of teaching status The various features included (e.g., frontier exploration width) Node saliency and progress stagnation The typical probability distribution of ); in performing likelihood In the calculation, based on the Naive Bayes assumption, it is assumed that the teaching state characteristics are... The various features included in the teaching strategy Given that they are independent under certain conditions, the likelihood is... It can be determined by the characteristics of the teaching status The likelihood is obtained by multiplying the conditional probabilities corresponding to each feature. The expression is as follows: ; in, Indicating teaching strategies When applicable, the breadth of frontier exploration The corresponding conditional probability; Indicating teaching strategies When applicable, node salience The corresponding conditional probability; Indicating teaching strategies When applicable, progress stagnation The corresponding conditional probability.

[0077] Based on the above formula, the values ​​of each teaching strategy in the teaching strategy library can be calculated separately. Characteristics of teaching status Posterior probability .

[0078] For ease of understanding, Table 1 will be used as an example here.

[0079] Table 1

[0080] As shown in Table 1, the assumed teaching strategies As a tiered prompting strategy, teaching status characteristics Including progress stagnation Then the likelihood ,Right now The value can vary depending on the number of user interaction rounds.

[0081] Furthermore, after calculating each teaching strategy in the teaching strategy library... Characteristics of teaching status Posterior probability Afterwards, the system can adjust the teaching strategy accordingly. Characteristics of teaching status The posterior probability is given for each teaching strategy in the teaching strategy base. Random sampling is performed to select the target teaching strategy to be implemented.

[0082] It should be noted that random sampling is used here instead of directly selecting the term with the highest posterior probability. This can avoid the system from producing a mechanical and predictable response pattern, thereby improving the naturalness of human-computer interaction.

[0083] The interactive teaching method provided in this application introduces a personalized interactive strategy selection algorithm based on mind maps. This algorithm transforms dynamic and structured mind map information into precise and reasonable decisions for the next teaching action through three stages: feature extraction, probability inference, and strategy selection, thereby realizing intelligent decision-making for the entire interactive teaching system.

[0084] In some embodiments, the real-time mind map includes at least two interactive nodes, and the teaching status features include frontier exploration width, node salience, and progress stagnation; wherein, frontier exploration width is used to characterize the user's thinking tendency; node salience is used to characterize the importance of the user's current interactive node; the current interactive node is the interactive node selected by the user in the real-time mind map; and progress stagnation is used to characterize the fluency of the user's problem-solving process.

[0085] In this embodiment, the real-time mind map is a directed acyclic graph that represents the user's real-time problem-solving logic path and can reflect the user's latest problem-solving ideas.

[0086] The real-time mind map includes at least two interactive nodes and at least one edge. An edge is used to connect the two interactive nodes. A node is used to represent a reasoning step in the user's problem-solving process, and an edge is used to represent the logical dependency between the two interactive nodes.

[0087] Teaching status characteristics It can reflect the user's current problem-solving progress and thinking patterns, as well as the characteristics of the teaching status. Including the breadth of frontier exploration Node saliency and progress stagnation .

[0088] Among them, the breadth of frontier exploration Used to characterize a user's thinking tendencies, it can measure whether a user's thinking tends to be "breadth-first" divergent exploration or "depth-first" linear progression.

[0089] Node salience It is used to characterize the importance of the user's current interaction node and can be used to determine whether the user is currently on the critical path to solving the problem.

[0090] The current interaction node is the interaction node selected by the user in the real-time mind map.

[0091] Stagnation of progress It is used to characterize the fluency of a user's problem-solving process and can be used to determine whether a user has hit a mental block.

[0092] In some embodiments, the frontier exploration width is determined based on the number of active leaf nodes and the maximum depth of the real-time mind graph. The number of active leaf nodes is the number of recently created interactive nodes without child nodes in the real-time mind graph. Node salience is determined based on the in-degree and semantic weight of the current interactive node. Progress stagnation is determined based on the user's interaction rounds.

[0093] Optionally, the breadth of frontier exploration The expression is as follows: ; in, This indicates the number of active leaf nodes. Active leaf nodes refer to interactive nodes in the real-time mind map that have recently been created or followed and have no child nodes. This indicates the maximum depth of the real-time mind map.

[0094] Optionally, node saliency The expression is as follows: ; in, Indicates the current interaction node; Represents interactive nodes The in-degree, where an in-degree greater than 1 indicates an interactive node. It is a logical convergence point that brings together multiple preliminary ideas; It is an adjustable hyperparameter; For interactive nodes Semantic weights, interaction nodes include policy information, interaction nodes The semantic weights can be determined by the large language model based on the interaction nodes. The strategy information is obtained through evaluation, and its value ranges from [0, 1], used to measure the interaction nodes. The criticality of the reasoning process.

[0095] Optionally, progress stagnation The number of interaction rounds between the user and the system can be obtained by counting the number of effective graph structure expansions since the previous round (i.e., the addition of interactive nodes to the real-time mind map).

[0096] In some embodiments, the real-time mind graph is a directed acyclic graph representing the user's real-time problem-solving logic path; wherein, the real-time mind graph includes at least two interactive nodes and at least one edge, an edge is used to connect the two interactive nodes, an interactive node is used to represent a reasoning step in the user's problem-solving process, and an edge is used to represent the logical dependency relationship between the two interactive nodes.

[0097] like Figure 6 As shown, in order to achieve accurate structured modeling of the user's problem-solving process, while taking into account the simplicity and flexibility of the graph model, the mind graph data model module defines a mind graph model. This model adopts a unified node data structure, which enables consistent representation of all types of thinking steps. At the same time, it uses structured fields to record the content of different cognitive levels, realizing a fine representation of thinking content.

[0098] Specifically, in this embodiment, the real-time mind map is a directed acyclic graph (DAG) that represents the user's real-time problem-solving logic path, denoted as G=(V,E).

[0099] The real-time mind graph G includes at least two interactive nodes and at least one edge, V represents the set of all interactive nodes of the real-time mind graph, and E represents the set of all edges of the real-time mind graph.

[0100] An edge is used to connect two interactive nodes. An interactive node represents a reasoning step in the user's problem-solving process, and an edge represents the logical dependency between the two interactive nodes.

[0101] For example, a directed edge from interaction node A to interaction node B indicates that the reasoning process of interaction node B depends on the conclusion of interaction node A.

[0102] In real-time mind maps, the connections between edges can contain a wealth of logical information.

[0103] Optionally, when an interactive node points to multiple child nodes, the edges of that interactive node are one-to-many connection edges.

[0104] Alternatively, a one-to-many connection edge can represent the following two cases: (1) Categorical discussion: Starting from a precondition, we can introduce a variety of situations that need to be discussed separately.

[0105] (2) Multi-path exploration: For the same problem, try multiple different problem-solving strategies.

[0106] Optionally, when multiple interactive nodes point to the same child node, the edges between the multiple interactive nodes are many-to-one connection edges.

[0107] Optionally, when multiple interactive nodes point to the same child node, it means that the establishment of the successor child node requires the conclusions of multiple predecessor nodes as a common premise.

[0108] In some embodiments, each interaction node includes strategy information, reasoning information, and result information; wherein, strategy information is used to describe the problem-solving strategy adopted by the corresponding reasoning step; reasoning information is used to describe the logical reasoning process of the corresponding reasoning step; and result information is used to describe the reasoning result of the corresponding reasoning step.

[0109] Specifically, all interactive nodes in the real-time mind map G adopt a unified type, defined as "reasoning step node". Each interactive node contains a set of structured fields to record complete information about the thinking steps.

[0110] Optionally, the structured fields include strategy information, reasoning information, and result information.

[0111] Among them, strategy information is used to describe the problem-solving strategy adopted by the reasoning steps corresponding to the interaction nodes, the macro-intention of the corresponding reasoning steps, or the methodology adopted.

[0112] For example, strategy information could be "prove using mathematical induction", "differentiate the function f(x) to analyze its monotonicity", or "try to simplify the integral using substitution".

[0113] By unifying the semantic functions of thought processes through strategy information description, the need to pre-define multiple fixed node types (such as "sub-problem nodes," "knowledge point nodes," etc.) can be avoided. For example, a step of "introducing a knowledge point" can be represented by an interactive node with the strategy "applying the Pythagorean theorem"; a step of "decomposing sub-problems" can be represented by an interactive node with the strategy "first prove that the auxiliary line AB is perpendicular to CD"; and an exploratory idea can be represented by an interactive node with the strategy "try to assume the conclusion is false in order to find contradictions." This design significantly enhances the flexibility and scalability of the mind map model, enabling it to represent diverse thought patterns through natural language descriptions, while simplifying the complexity of the underlying data model.

[0114] Reasoning information is used to describe the specific logical reasoning process of the reasoning steps corresponding to the interaction node, including mathematical calculations, logical deductions, or textual arguments.

[0115] For example, the reasoning information could be: "When n=1, the left side = 1, the right side = 1, and the equation holds. Assuming the equation holds when n=k, then when n=k+1, another possibility can be deduced."

[0116] The results information is used to describe the reasoning results of the reasoning steps corresponding to the interaction nodes, and to summarize the conclusions reached or obstacles encountered in the current step.

[0117] For example, the result information could be "proved that the conclusion holds when n=1", "proved that the monotonically increasing interval of the function is (a, b)", "this path is not feasible, and the integration becomes more complicated", etc.

[0118] Optionally, in addition to the strategy information, reasoning information, and result information mentioned above, each interaction node also includes the following basic attribute information, which is used to define the unique identifier and connection relationship of the interaction node in the real-time mind map G: (1) Node ID (node_id): A unique identifier for the interacting node.

[0119] (2) Parent node list (parent_ids): A list of IDs of the parent nodes of the currently interacting node.

[0120] (3) Child node list (child_ids): A list of IDs of the child nodes pointed to by the current node.

[0121] Optionally, the mind map data model module is also used to define the mind map update interface.

[0122] Specifically, such as Figure 6 As shown, to support the large language model in performing programmatic and precise updates to mind maps, the mind map data model module also defines a set of application programming interfaces (APIs), namely the mind map update interface. The large language model constructs and dynamically updates dynamic mind maps by calling the mind map update interface.

[0123] Optionally, the update interface functions of the mind map update interface include the following categories: (1) The create_node(content) function is used to create a new interactive node. The content parameter is structured data containing the initial values ​​of fields such as strategy information, inference information and result information. This function can return the node ID of the newly created interactive node.

[0124] (2) The update_node(node_id, new_content) function is used to update the content of an interactive node with a specified ID. For example, it can populate previously empty inference information fields or correct erroneous calculation results.

[0125] (3) The create_edge(source_node_id, target_node_id) function is used to create a directed edge between the source node and the target node.

[0126] (4) The get_graph_state function is used to obtain the current state of the entire mind graph. It is usually returned in a structured format such as JSON so that the large language model can obtain the complete context of the problem-solving process before making decisions and generating responses.

[0127] In addition, to ensure that the large language model can accurately understand the current problem-solving context during the interaction process and generate appropriate mind map operation instructions and teaching response text accordingly, this embodiment also needs to design a context management module. This module is responsible for dynamically constructing a structured input context (Prompt) for each call of the large language model. This input context integrates all the information required for the large language model to make decisions.

[0128] Optionally, the input context includes the following three types of information: (1) System-level instructions: A set of meta-instructions set at the start of a session and used throughout the session. These meta-instructions define the core roles (e.g., "math tutor"), objectives (to guide users to think rather than give them answers directly), and behavioral guidelines that must be followed in the large language model.

[0129] (2) Contextual information: It contains the complete background of the current interaction, mainly including the global state of the real-time mind map (the complete data structure of the real-time mind map, usually given in a structured format such as JSON, covering the information of all interactive nodes and their connection relationships), historical dialogue (recent user and system interaction records, used to maintain the continuity of the dialogue) and input text (i.e. the user's latest input text and the node ID of the interactive node selected by the user on the visualization interface).

[0130] (3) Task instructions: Task instructions are mainly used to provide clear execution goals based on the current stage of the process. These instructions are mainly divided into two categories: First, instructions for updating mind maps, which can provide detailed definitions of available tools (APIs), such as the create_node(content) function and the update_node(node_id, new_content) function. These definitions follow the standard tool calling format, clearly informing the large language model of its procedural operation capabilities. Its task is to decide which interfaces to call to update the mind map. Second, instructions for generating teaching response text, which can provide teaching strategy instructions determined by the personalized interaction strategy selection algorithm based on mind maps, such as "execute the Socratic questioning strategy" or "provide a hierarchical prompting strategy". In this case, the task of the large language model is to generate a guiding teaching response text based on the instruction.

[0131] Through the above mechanism, the context management module can organically combine system-level instructions, contextual information, and task instructions. Based on this, the large language model can not only refer to the complete problem-solving process and historical dialogues, but also accurately respond to the user's real-time input and strictly follow the core task instructions of the current stage, thereby driving the orderly operation of the entire teaching interaction process.

[0132] Compared with the prior art, the interactive teaching method provided in this application has at least the following technical advantages: (1) Provides real-time and intuitive cognitive assistance, significantly reducing the cognitive burden on users: The dynamic mind map proposed in this application can structure the user's scattered thoughts in real time during the interaction, clearly present the logical dependencies, branches and convergences between reasoning steps, so that users can grasp the problem-solving context and current state at a glance, avoid users getting lost in the complex reasoning chain, and overcome the cognitive burden of users having to reconstruct their thoughts from lengthy linear dialogues in the prior art.

[0133] (2) More precise teaching guidance is achieved, which can effectively adapt to the nonlinearity and diversity of user thinking: The personalized interactive strategy selection algorithm based on mind map proposed in this application describes the user's real-time thinking state by analyzing the structured mind map rather than the nonlinear historical dialogue, so that the system can accurately identify the branching, convergence and switching of the user's thinking, effectively solving the defect of existing algorithms that are difficult to model nonlinear thinking processes; at the same time, the method guides students according to their actual thinking, overcoming the closed nature of relying on preset answers in the existing technology.

[0134] (3) It realizes flexible and dynamic problem-solving process modeling, which can reflect the user's higher-order thinking ability: The mind map model proposed in this application adopts a unified and flexible reasoning step node, gets rid of the dependence on predefined knowledge points, and can accurately reflect the user's higher-order dynamic thinking ability in analysis, exploration and other aspects; In addition, the mind map model realizes the dynamic growth of the mind map in the interaction through the update interface, which solves the problem that the existing graph modeling method is lagging behind and can only be used for post-diagnosis, and truly realizes real-time assistance in the problem-solving process.

[0135] To facilitate understanding, a specific example of an interactive teaching method is provided here for reference.

[0136] like Figure 7 As shown, suppose the problem input by the user is: Given And inequalities If the condition is always true, then find the real number. The range of values ​​for .

[0137] The user's general solution process is as follows: (1) Sequential reasoning: The user begins to attempt to solve the problem, first determining the domain of the problem to be solved. (Interaction node N1), and analyzed The symbol (interaction node N2) clearly records the above problem-solving process in the mind map.

[0138] (2) Node branching: Based on the results of interaction node N2, users naturally engage in classification discussions, which include three cases: , and In the mind map, interaction node N2 is connected to interaction nodes N3, N4, and N5, which represent the three cases.

[0139] (3) Failure and backtracking: The user attempts to continue reasoning based on interaction node N4 (interaction node N6), but after thinking it over, the user finds that due to The symbols change periodically, making the situation very complex and difficult to analyze.

[0140] (4) System guidance and node convergence: When the system detects that the user has paused the solution, it will provide a prompt. Inspired by the system prompt, the user will try a new approach (interaction node N7). After comprehensively considering the conclusions of interaction nodes N4 and N5, the user will use continuity and boundary points to analyze and solve the problem. In the mind map, interaction nodes N4 and N5 are both connected to interaction node N7.

[0141] (5) Continue reasoning and complete the solution: Under the guidance of the system, the user discovers It was only a necessary condition, so the verification continued. This is also a sufficient condition (interaction node N8). Combining interaction nodes N7 and N8, the user finally completes the solution. The answer to this question is... .

[0142] This example shows that a user's actual thinking process naturally involves non-linear elements such as branching, backtracking, and convergence. The mind map provided in this application embodiment can concisely and clearly represent the user's complex thinking structure, helping the user to clarify their thoughts. The system can also use the mind map to more accurately analyze the user's thinking state and provide better teaching guidance.

[0143] This application also provides an interactive teaching device. Please refer to [link / reference]. Figure 8 , Figure 8 This is a schematic diagram of the structure of the interactive teaching device provided in an embodiment of this application. In this embodiment, the interactive teaching device includes an initialization module 810, a receiving module 820, a mind map updating module 830, a teaching strategy selection module 840, a generation module 850, and an interactive teaching module 860.

[0144] The initialization module 810 is used to create a dynamic mind map for the user based on the unanswered question input by the user.

[0145] Dynamic mind maps are used to represent the user's problem-solving logic path.

[0146] The receiving module 820 is used to receive user input text.

[0147] The input text is a description of the user's thought process in solving the problem to be solved.

[0148] The mind map update module 830 is used to input dynamic mind maps, input text, and the user's historical dialogues into the large language model to update the mind map and obtain the real-time mind map output by the large language model.

[0149] The instructional strategy selection module 840 is used to select a target instructional strategy from multiple instructional strategies based on real-time mind maps.

[0150] The generation module 850 is used to input the target teaching strategy, input text, historical dialogue and real-time mind map into the large language model to obtain the teaching response text output by the large language model.

[0151] The interactive teaching module 860 is used to treat the real-time mind map as a dynamic mind map and return to the step of receiving the user's input text until the user successfully solves the problem.

[0152] In some embodiments, selecting a target teaching strategy from multiple teaching strategies based on a real-time mind map includes: extracting features from the real-time mind map to generate teaching state features of the real-time mind map; calculating the posterior probability of each teaching strategy in the teaching strategy library under the teaching state features based on Bayes' theorem; and randomly sampling each teaching strategy in the teaching strategy library based on the posterior probability of each teaching strategy under the teaching state features to select the target teaching strategy.

[0153] In some embodiments, the real-time mind map includes at least two interactive nodes, and the teaching status features include frontier exploration width, node salience, and progress stagnation; wherein, frontier exploration width is used to characterize the user's thinking tendency; node salience is used to characterize the importance of the user's current interactive node; the current interactive node is the interactive node selected by the user in the real-time mind map; and progress stagnation is used to characterize the fluency of the user's problem-solving process.

[0154] In some embodiments, the frontier exploration width is determined based on the number of active leaf nodes and the maximum depth of the real-time mind graph. The number of active leaf nodes is the number of recently created interactive nodes without child nodes in the real-time mind graph. Node salience is determined based on the in-degree and semantic weight of the current interactive node. Progress stagnation is determined based on the user's interaction rounds.

[0155] In some embodiments, the real-time mind graph is a directed acyclic graph representing the user's real-time problem-solving logic path; wherein, the real-time mind graph includes at least two interactive nodes and at least one edge, an edge is used to connect the two interactive nodes, an interactive node is used to represent a reasoning step in the user's problem-solving process, and an edge is used to represent the logical dependency relationship between the two interactive nodes.

[0156] In some embodiments, each interaction node includes strategy information, reasoning information, and result information; wherein, strategy information is used to describe the problem-solving strategy adopted by the corresponding reasoning step; reasoning information is used to describe the logical reasoning process of the corresponding reasoning step; and result information is used to describe the reasoning result of the corresponding reasoning step.

[0157] This application also provides an electronic device. Figure 9 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application, such as... Figure 9 As shown, the electronic device may include a processor 910, a communications interface 920, a memory 930, and a communication bus 940. The processor 910, communications interface 920, and memory 930 communicate with each other via the communication bus 940. The processor 910 can call logical instructions from the memory 930 to execute interactive teaching methods.

[0158] Furthermore, the logical instructions in the aforementioned memory 930 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0159] This application also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the interactive teaching methods provided by the above methods.

[0160] This application also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can perform the interactive teaching methods provided by the above methods.

[0161] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0162] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0163] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. An interactive teaching method, characterized in that, include: Create a dynamic mind map for the user based on the unanswered question input by the user; The dynamic mind map is used to represent the user's problem-solving logic path; Receive the user's input text; the input text is a text describing the user's approach to solving the problem to be solved; The dynamic mind map, the input text, and the user's historical dialogue are input into the large language model to update the mind map and obtain the real-time mind map output by the large language model. Based on the real-time mind map, a target teaching strategy is selected from multiple teaching strategies; The target teaching strategy, the input text, the historical dialogue, and the real-time mind map are input into the large language model to obtain the teaching response text output by the large language model. The real-time mind map is used as the dynamic mind map, and the process returns to the step of receiving the user's input text until the user successfully solves the problem.

2. The interactive teaching method according to claim 1, characterized in that, The selection of a target teaching strategy from multiple teaching strategies based on the real-time mind map includes: Feature extraction is performed on the real-time mind map to generate the teaching status features of the real-time mind map; Based on Bayes' theorem, the posterior probability of each teaching strategy in the teaching strategy library under the teaching state features is calculated respectively. Based on the posterior probability of each teaching strategy under the teaching state features, each teaching strategy in the teaching strategy library is randomly sampled to select the target teaching strategy.

3. The interactive teaching method according to claim 2, characterized in that, The real-time mind map includes at least two interactive nodes, and the teaching status characteristics include the breadth of frontier exploration, node salience, and progress stagnation. The breadth of frontier exploration is used to characterize the user's thinking tendency; The node saliency is used to characterize the importance of the user's current interaction node; the current interaction node is the interaction node selected by the user in the real-time mind map. The progress stagnation level is used to characterize the smoothness of the user's problem-solving process.

4. The interactive teaching method according to claim 3, characterized in that, The frontier exploration width is determined based on the number of active leaf nodes and the maximum depth of the real-time mind graph, where the number of active leaf nodes is the number of recently created interactive nodes without child nodes in the real-time mind graph; The saliency of a node is determined based on the in-degree of the currently interacting node and the semantic weight of the currently interacting node. The progress stagnation level is determined based on the number of user interaction rounds.

5. The interactive teaching method according to claim 1, characterized in that, The real-time mind map is a directed acyclic graph that represents the user's real-time problem-solving logic path; The real-time mind map includes at least two interactive nodes and at least one edge. One edge connects the two interactive nodes, one interactive node represents a reasoning step in the user's problem-solving process, and one edge represents the logical dependency between the two interactive nodes.

6. The interactive teaching method according to claim 5, characterized in that, Each of the aforementioned interaction nodes includes strategy information, reasoning information, and result information; The strategy information is used to describe the problem-solving strategy adopted in the corresponding reasoning step; The reasoning information is used to describe the logical reasoning process of the corresponding reasoning step; The result information is used to describe the reasoning result of the corresponding reasoning step.

7. An interactive teaching device, characterized in that, include: An initialization module is used to create a dynamic mind map for the user based on the unanswered question input by the user. The dynamic mind map is used to represent the user's problem-solving logic path; A receiving module is used to receive the user's input text; the input text is a text describing the user's approach to solving the problem to be solved; The mind map update module is used to input the dynamic mind map, the input text, and the user's historical dialogue into the large language model to update the mind map and obtain the real-time mind map output by the large language model. The teaching strategy selection module is used to select a target teaching strategy from multiple teaching strategies based on the real-time mind map. The generation module is used to input the target teaching strategy, the input text, the historical dialogue and the real-time mind map into the large language model to obtain the teaching response text output by the large language model; The interactive teaching module is used to use the real-time mind map as the dynamic mind map and return to the step of receiving the user's input text until the user successfully solves the question to be answered.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the interactive teaching method as described in any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the interactive teaching method as described in any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the interactive teaching method as described in any one of claims 1 to 6.

Images