Experiment process control and interactive guidance system, method, electronic device and program product

Abstract

The application provides an experimental process control and interactive guidance system, method, electronic equipment and program product. The system accurately anchors the current physical experiment step of the user through a multi-modal intention analysis unit, solves the context misplacement and cross-step illusion problems that occur in the guidance process of a general large language model from the root, and ensures the accuracy and process synchronization of the generated guidance content. The system integrates a preposed safety and strategy control unit, builds a mandatory safety constraint mechanism, effectively avoids the risk of dangerous operation suggestions that may be generated by a general model through real-time risk detection and fuse blocking, and significantly improves the safety of the experimental teaching scene. At the same time, the system uses an improved Bayesian knowledge tracking algorithm to continuously quantify and update the user's cognitive state through a dynamic cognitive modeling unit, and adaptively selects different tutoring strategy templates based on this, thereby realizing personalized and hierarchical tutoring.

Description

Technical Field

[0001] This application relates to the field of large language model technology, and more specifically, to an experimental process control and interactive guidance system, method, electronic device and program product. Background Technology

[0002] In STEM experimental teaching scenarios, students often encounter problems such as confusion about operating procedures, misunderstanding of experimental phenomena, or unclear understanding of underlying principles when conducting experiments in physics, chemistry, and electronics. Traditional auxiliary methods mainly rely on on-site guidance from teachers or teaching assistants. This approach has limitations such as large response delays, limited human resources, and difficulty in providing personalized tutoring, and cannot meet the needs of large-scale, high-concurrency experimental teaching.

[0003] With the rapid development of generative artificial intelligence (AIGC) technology, intelligent question-answering systems based on large language models (such as ChatGPT and GPT-4) have been tested and applied in educational assistance scenarios. These general-purpose models can generate fluent and coherent text responses based on massive corpora, which alleviates the pressure of human tutoring to some extent.

[0004] General-purpose models typically generate responses based solely on the text history of the current dialogue, failing to proactively or accurately determine the specific physical step the student is currently in within a standard experimental procedure. For example, a student might be performing step 2 of circuit connection, but the model might semantically link it to step 5 of data processing. This misalignment of physical context can easily lead to generated guidance that is premature, delayed, or completely irrelevant to the current operation, creating a "cross-step illusion" that severely disrupts the experimental process and can even cause operational errors.

[0005] Experiments, especially those involving high-voltage electricity, flammable and explosive chemicals, or high-temperature and high-pressure equipment, are potentially dangerous. General-purpose language models are not trained with safety rule knowledge bases or real-time risk assessment logic specific to particular experimental scenarios. When users describe or inquire about dangerous operations, the model may generate seemingly reasonable but generic responses lacking necessary safety warnings or containing dangerous advice, posing a significant risk of inducing users to perform unsafe operations.

[0006] Effective tutoring should adhere to the principle of individualized instruction, dynamically adjusting the depth and method of tutoring strategies based on students' current knowledge levels. General language models typically operate on a conversational basis, lacking the ability to continuously model and memorize users' long-term, cross-conversational knowledge mastery. Each response is essentially independent, failing to achieve consistency and adaptive adjustment of tutoring strategies based on students' historical performance curves (such as changes in the probability of mastering a particular knowledge point), thus hindering truly personalized, progressive guidance. Summary of the Invention

[0007] The purpose of this application is to provide an experimental process control and interactive guidance system, method, electronic device and program product to solve the problems that the prior art is unable to meet the requirements of real-time and accurate context awareness, mandatory safety risk prevention and control, and personalized strategy generation based on continuous cognitive assessment.

[0008] This application provides an experimental process control and interactive guidance system, including:

[0009] The multimodal intent parsing unit is used to receive and parse text and / or image input from the user, and, in conjunction with a preset experimental flowchart, determine the current physical experimental step in which the user is.

[0010] The dynamic cognitive modeling unit, connected to the multimodal intent parsing unit, is used to run an improved Bayesian knowledge tracing algorithm based on the user's historical interaction data, calculate and dynamically update the probability value of the user's cognitive mastery of the knowledge points related to the current step;

[0011] The security and policy control unit is connected to the multimodal intent parsing unit and the dynamic cognitive modeling unit, respectively, and is used to: perform risk detection on user input based on a preset keyword library and / or image danger feature library; when the risk value is determined to exceed a preset threshold, trigger a security circuit breaker mechanism to block the generation of a regular response; when the risk value is determined to be less than the preset threshold, select a target policy template from multiple preset tutoring policy templates according to the numerical range of the cognitive mastery probability value output by the dynamic cognitive modeling unit.

[0012] The generative interaction unit, connected to the safety and policy control unit, is used to generate a guiding response for the user by calling a large language model based on the target policy template and the knowledge base context related to the current physical experiment steps when the risk value is determined to be no more than a preset threshold.

[0013] In the aforementioned technical solution, the system accurately anchors the user's current physical experiment steps through a multimodal intent parsing unit, fundamentally solving the contextual misalignment and cross-step illusion problems that occur in the guidance process of general-purpose large language models, ensuring the accuracy of generated guidance content and the synchronization of the process. The system integrates a pre-positioned safety and policy control unit, constructing a mandatory safety constraint mechanism. Through real-time risk detection and circuit breaker blocking, it effectively avoids the risk of dangerous operation suggestions that general-purpose models may generate, significantly improving the safety of experimental teaching scenarios. Simultaneously, through a dynamic cognitive modeling unit, the system continuously quantifies and updates the user's cognitive state using an improved Bayesian knowledge tracing algorithm, and adaptively selects different tutoring strategy templates based on this, thereby achieving personalized, tiered tutoring. This transforms vague teaching experience into a quantifiable and reproducible intelligent decision-making process, greatly optimizing the allocation efficiency of teaching resources and the student learning experience.

[0014] In some alternative implementations, the Bayesian knowledge tracing algorithm is improved by incorporating the user's rating results in the experimental operation dimension into the update calculation of the cognitive mastery probability value.

[0015] In the above technical solution, by integrating the user's rating results in the experimental operation dimension into the cognitive mastery probability value update calculation of the improved Bayesian knowledge tracing algorithm, this system achieves a more comprehensive and accurate quantitative assessment of the user's cognitive state. It breaks through the single-dimensional limitation of the traditional BKT algorithm, which only relies on the correctness of the answers, and incorporates practical indicators such as the correctness, standardization, and proficiency of the experimental operation into the evaluation, thereby constructing a multi-dimensional cognitive assessment model that combines theoretical understanding and practical operation.

[0016] In some alternative implementations, the improved Bayesian knowledge tracing algorithm also introduces a forgetting function to decay the probability value of knowledge acquisition based on time intervals.

[0017] In the aforementioned technical solution, a forgetting function is introduced into the improved Bayesian knowledge tracing algorithm to decay and correct the probability value of knowledge mastery based on time intervals. This system simulates the objective forgetting law of human memory, making the evaluation model of the dynamic cognitive modeling unit closer to real teaching situations. It avoids the overconfidence in cognitive state caused by traditional models that only accumulate positive feedback, and can more realistically reflect the degree of forgetting of specific knowledge points by students after a long period of no practice. By introducing a decay mechanism in the time dimension, the system can dynamically and scientifically adjust tutoring strategies. For example, when it detects that the probability value of a student's knowledge point has decayed to a lower range due to forgetting, it automatically triggers review prompts or reduces the difficulty of guidance, thereby providing more personalized tutoring that conforms to cognitive laws, is more humane, and has long-term effectiveness, improving the system's adaptive teaching capabilities and user experience.

[0018] In some optional implementations, the numerical range of the probability value of cognitive mastery is used for strategy routing through a preset decision tree structure, wherein the numerical range is at least divided into a low probability range corresponding to novices, a medium probability range corresponding to proficient users, and a high probability range corresponding to expert users.

[0019] In the above technical solution, through a layered and structured strategy distribution mechanism, the system can provide novice users with more basic explanations and error correction guidance, skilled users with heuristic key feature prompts, and expert users with deeper counterfactual reasoning guidance. This ensures that the difficulty and method of the tutoring content are always highly adapted to the user's current actual ability level, significantly improves the personalization of teaching and guidance efficiency, and makes the entire tutoring decision-making process highly transparent, standardized, and reproducible.

[0020] In some optional implementations, when the security circuit breaker is triggered, a preset hard-coded security warning message is directly output to the user.

[0021] In the above technical solution, the safety circuit breaker mechanism bypasses all potential large language model generation paths that contain uncertainties, achieves the highest priority security protection, provides immediate, clear, and risk-free warnings, effectively interrupts potentially dangerous experimental processes, minimizes the security risks caused by improper AI responses, and fundamentally enhances the overall security of the system.

[0022] In some alternative implementations, the generative interaction unit is also configured to retrieve knowledge base context related to the current step and its predecessor steps from the vector database based on the current physical experiment step, and to mask information of subsequent steps.

[0023] The above technical solution fundamentally prevents cross-step information illusions caused by knowledge base information leakage, ensuring that the large language model will not explain the phenomena or problems of the current step based on the knowledge of subsequent experimental steps, thereby completely eliminating the risk of spoilers and ensuring the natural exploration and logical coherence of the learning process.

[0024] In some alternative implementations, the multimodal intent parsing unit determines the current physical experiment step the user is in using image recognition technology and / or system background operation logs.

[0025] This application provides an experimental process control and interactive guidance method, including:

[0026] Receive and parse text and / or image input from the user, and combine it with a preset experimental flowchart to determine the current physical experimental step the user is in;

[0027] Based on the user's historical interaction data, run the improved Bayesian knowledge tracing algorithm to calculate and dynamically update the probability value of the user's knowledge mastery of the relevant knowledge points in the current step;

[0028] Based on a pre-set keyword library and / or image hazard feature library, risk detection is performed on user input; when the risk value is determined to exceed a pre-set threshold, a safety circuit breaker mechanism is triggered to block the generation of a regular response; when the risk value is determined to be within the pre-set threshold, a target strategy template is selected from multiple pre-set tutoring strategy templates based on the numerical range of the cognitive mastery probability value.

[0029] When the risk value is determined to be within a preset threshold, a large language model is invoked to generate a guiding response for the user based on the target strategy template and the knowledge base context related to the current physical experiment steps.

[0030] An electronic device provided in this application includes a processor and a memory, wherein the memory stores machine-readable instructions executable by the processor, and the machine-readable instructions, when executed by the processor, perform any of the methods described above.

[0031] This application provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the method described above. Attached Figure Description

[0032] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0033] Figure 1 A functional module diagram of an experimental process control and interactive guidance system provided in this application embodiment;

[0034] Figure 2 A flowchart illustrating the steps of an experimental process control and interactive guidance method provided in this application embodiment. Detailed Implementation

[0035] The technical solutions in the embodiments of this application will now be described with reference to the accompanying drawings.

[0036] Please refer to Figure 1 , Figure 1 A functional block diagram of an experimental process control and interactive guidance system provided in this application embodiment includes:

[0037] The multimodal intent parsing unit is used to receive and parse text and / or image input from the user, and, in conjunction with a preset experimental flowchart, determine the current physical experimental step in which the user is.

[0038] The dynamic cognitive modeling unit, connected to the multimodal intent parsing unit, is used to run an improved Bayesian knowledge tracing algorithm based on the user's historical interaction data, calculate and dynamically update the probability value of the user's cognitive mastery of the knowledge points related to the current step;

[0039] The security and policy control unit is connected to the multimodal intent parsing unit and the dynamic cognitive modeling unit, respectively, and is used to: perform risk detection on user input based on a preset keyword library and / or image danger feature library; when the risk value is determined to exceed a preset threshold, trigger a security circuit breaker mechanism to block the generation of a regular response; when the risk value is determined to be less than the preset threshold, select a target policy template from multiple preset tutoring policy templates according to the numerical range of the cognitive mastery probability value output by the dynamic cognitive modeling unit.

[0040] The generative interaction unit, connected to the safety and policy control unit, is used to generate a guiding response for the user by calling a large language model based on the target policy template and the knowledge base context related to the current physical experiment steps when the risk value is determined to be no more than a preset threshold.

[0041] In the aforementioned technical solution, the system accurately anchors the user's current physical experiment steps through a multimodal intent parsing unit, fundamentally solving the contextual misalignment and cross-step illusion problems that occur in the guidance process of general-purpose large language models, ensuring the accuracy of generated guidance content and the synchronization of the process. The system integrates a pre-positioned safety and policy control unit, constructing a mandatory safety constraint mechanism. Through real-time risk detection and circuit breaker blocking, it effectively avoids the risk of dangerous operation suggestions that general-purpose models may generate, significantly improving the safety of experimental teaching scenarios. Simultaneously, through a dynamic cognitive modeling unit, the system continuously quantifies and updates the user's cognitive state using an improved Bayesian knowledge tracing algorithm, and adaptively selects different tutoring strategy templates based on this, thereby achieving personalized, tiered tutoring. This transforms vague teaching experience into a quantifiable and reproducible intelligent decision-making process, greatly optimizing the allocation efficiency of teaching resources and the student learning experience.

[0042] This application embodiment constructs a closed-loop control system of "perception-decision-execution" and adopts a dual-stream state perception model: the physical stream determines which state node the student is in in the experimental flowchart through the experimental manual anchor point and multimodal input; the cognitive stream uses an improved Bayesian knowledge tracing algorithm to maintain a dynamically updated "cognitive state vector table" to record the probability of the student's mastery of each knowledge point.

[0043] In some alternative implementations, the Bayesian knowledge tracing algorithm is improved by incorporating the user's rating results in the experimental operation dimension into the update calculation of the cognitive mastery probability value.

[0044] In the above technical solution, by integrating the user's rating results in the experimental operation dimension into the cognitive mastery probability value update calculation of the improved Bayesian knowledge tracing algorithm, this system achieves a more comprehensive and accurate quantitative assessment of the user's cognitive state. It breaks through the single-dimensional limitation of the traditional BKT algorithm, which only relies on the correctness of the answers, and incorporates practical indicators such as the correctness, standardization, and proficiency of the experimental operation into the evaluation, thereby constructing a multi-dimensional cognitive assessment model that combines theoretical understanding and practical operation.

[0045] In some alternative implementations, the improved Bayesian knowledge tracing algorithm also introduces a forgetting function to decay the probability value of knowledge acquisition based on time intervals.

[0046] In the aforementioned technical solution, a forgetting function is introduced into the improved Bayesian knowledge tracing algorithm to decay and correct the probability value of knowledge mastery based on time intervals. This system simulates the objective forgetting law of human memory, making the evaluation model of the dynamic cognitive modeling unit closer to real teaching situations. It avoids the overconfidence in cognitive state caused by traditional models that only accumulate positive feedback, and can more realistically reflect the degree of forgetting of specific knowledge points by students after a long period of no practice. By introducing a decay mechanism in the time dimension, the system can dynamically and scientifically adjust tutoring strategies. For example, when it detects that the probability value of a student's knowledge point has decayed to a lower range due to forgetting, it automatically triggers review prompts or reduces the difficulty of guidance, thereby providing more personalized tutoring that conforms to cognitive laws, is more humane, and has long-term effectiveness, improving the system's adaptive teaching capabilities and user experience.

[0047] Specifically, the core calculation formula for the improved Bayesian knowledge tracing algorithm is as follows:

[0048] The probability update formula for mastering knowledge points at time t is: ;

[0049] The formula for updating the evaluation criteria for the integrated experimental operation is: ;

[0050] The modified formula for introducing the forgetting function is: ; in ; This represents the probability value of a user's understanding and mastery of the current knowledge point at time t; This represents the mastery probability updated based on user interaction feedback at time t-1; This represents the state transition probability (i.e., the learning probability), which is the inherent probability that a user will transition from an unmastered state to a mastered state. This represents the comprehensive interactive feedback score at time t-1; α and β are the weighting coefficients of the question-and-answer score and the operation score, respectively, used to balance the proportion of theoretical understanding and practical operation. This indicates the user's rating results in question-and-answer interaction dimensions (such as the correctness of answers and the quality of text interaction); This indicates the user's rating results in experimental operation dimensions (such as operation correctness, standardization, and proficiency); λ represents the final probability value of cognitive mastery after time decay correction; F(Δt) represents the Ebbinghaus forgetting function, used to simulate the objective forgetting law of human memory; Δt represents the time interval between the current moment and the last interaction or learning of this knowledge point; λ represents the forgetting decay coefficient, reflecting the forgetting rate of a specific knowledge point or a specific user.

[0051] In some optional implementations, the numerical range of the probability value of cognitive mastery is used for strategy routing through a preset decision tree structure, wherein the numerical range is at least divided into a low probability range corresponding to novices, a medium probability range corresponding to proficient users, and a high probability range corresponding to expert users.

[0052] In the above technical solution, through a layered and structured strategy distribution mechanism, the system can provide novice users with more basic explanations and error correction guidance, skilled users with heuristic key feature prompts, and expert users with deeper counterfactual reasoning guidance. This ensures that the difficulty and method of the tutoring content are always highly adapted to the user's current actual ability level, significantly improves the personalization of teaching and guidance efficiency, and makes the entire tutoring decision-making process highly transparent, standardized, and reproducible.

[0053] In this embodiment, the deterministic policy routing algorithm distributes users based on the specific value of the cognitive mastery probability P: when P < 0.3, the user is identified as a novice and routed to the error correction and definition explanation template; when 0.3 ≤ P < 0.7, the user is identified as an expert and routed to the key feature prompt template (or heuristic prompt strategy); when P ≥ 0.7, the user is identified as an expert and routed to the counterfactual reasoning template.

[0054] In some optional implementations, when the security circuit breaker is triggered, a preset hard-coded security warning message is directly output to the user.

[0055] In the above technical solution, the safety circuit breaker mechanism bypasses all potential large language model generation paths that contain uncertainties, achieves the highest priority security protection, provides immediate, clear, and risk-free warnings, effectively interrupts potentially dangerous experimental processes, minimizes the security risks caused by improper AI responses, and fundamentally enhances the overall security of the system.

[0056] In this embodiment, the safety circuit breaker mechanism serves as a high-priority pre-safety filter, calculated based on risk entropy. The preset keyword library includes, but is not limited to, dangerous words such as "spark," "heat," and "short circuit." Once triggered, the generation path of the large language model is immediately cut off, forcibly switching to a hard-coded blocking warning mode.

[0057] In some alternative implementations, the generative interaction unit is also configured to retrieve knowledge base context related to the current step and its predecessor steps from the vector database based on the current physical experiment step, and to mask information of subsequent steps.

[0058] The above technical solution fundamentally prevents cross-step information illusions caused by knowledge base information leakage, ensuring that the large language model will not explain the phenomena or problems of the current step based on the knowledge of subsequent experimental steps, thereby completely eliminating the risk of spoilers and ensuring the natural exploration and logical coherence of the learning process.

[0059] Specifically, the context retrieval algorithm based on experimental process anchoring includes: defining an experimental flowchart G=(V,E), and obtaining the user's current location. Construct the retrieval query vector Q = Embed(user question) + Embed(steps) (Key parameters).

[0060] Where G=(V,E) represents the directed graph model of the experimental process; V represents the set of nodes of the experimental steps (e.g., connecting circuits, adjusting oscilloscopes, etc.); E represents the set of edges that represent the sequence or logical connection between the experimental steps. This indicates the current physical experiment step node where the user is; Q represents the comprehensive query vector used for similarity retrieval in the vector database; Embed() represents the vectorized embedding function for text or parameters, used to convert text into a high-dimensional vector representation.

[0061] In some optional implementations, the multimodal intent parsing unit determines the user's current physical experiment step using image recognition technology and / or system background operation logs. During this process, the multimodal intent parsing unit performs semantic disambiguation in conjunction with the current experiment step pointer to ensure accurate anchoring of the physical flow state.

[0062] Please refer to Figure 2 , Figure 2 A flowchart of an experimental process control and interactive guidance method provided in this application embodiment includes:

[0063] Step 1: Receive and parse text and / or image input from the user, and determine the current physical experiment step the user is in by combining it with the preset experimental flowchart;

[0064] Step 2: Based on the user's historical interaction data, run the improved Bayesian knowledge tracing algorithm to calculate and dynamically update the probability value of the user's knowledge mastery of the relevant knowledge points in the current step;

[0065] Step 3: Based on the preset keyword library and / or image hazard feature library, perform risk detection on user input; when the risk value is determined to exceed the preset threshold, trigger the safety circuit breaker mechanism to block the generation of regular responses; when the risk value is determined to be within the preset threshold, select a target strategy template from multiple preset tutoring strategy templates according to the numerical range of the cognitive mastery probability value.

[0066] Step 4: When the risk value is determined to be no more than the preset threshold, based on the target strategy template and the knowledge base context related to the current physical experiment step, the large language model is invoked to generate a guiding response for the user.

[0067] In one specific example, a student is conducting an RLC series resonance experiment. The current operation step is "Experimental Step 3: Measuring Amplitude-Frequency Characteristics". The student uploads a screenshot of the distorted waveform observed on an oscilloscope and asks, "Why is the top of the waveform flat?"

[0068] Step 1: Multimodal intent parsing and physical state localization.

[0069] Input: The text "Why is the top of the waveform flat?" uploaded by the user and the waveform image from the oscilloscope.

[0070] Parsing process: The multimodal intent parsing unit simultaneously receives text and image input.

[0071] Text parsing: semantic understanding of the question.

[0072] Image analysis and state anchoring: The system analyzes waveform features (such as top flattening) through image recognition technology and combines them with the student operation logs recorded in the system background (for example, the student has just completed the second step of the experiment, "Setting the signal source", and entered the third step of the experiment). The system then determines that the student is currently in the third step of the preset "experiment flowchart", namely the "measuring amplitude and frequency characteristics" step.

[0073] Output: Determines the user's current physical experiment step as Experiment Step 3.

[0074] Step 2: Dynamic cognitive modeling and state assessment.

[0075] Input: The current experimental step three determined in step 1, and historical data related to the knowledge points of this step (such as "signal distortion" and "clipping phenomenon") retrieved from the user's historical interaction database.

[0076] Algorithm execution: The improved Bayesian Knowledge Trace (BKT) algorithm is used by the dynamic cognitive modeling unit to calculate the probability value of the student's cognitive mastery of the relevant knowledge points in the current step.

[0077] Output: The system queries the student's historical probability of knowing the knowledge point "clipping distortion," assuming the current probability value after forgetting correction is 0.45. The system determines that the student is in the proficiency range (e.g., preset ranges are: Novice [0, 0.3), Proficient [0.3, 0.7), Expert [0.7, 1.0]).

[0078] Step 3: Security Detection and Policy Routing Decision

[0079] Input: Text and images input by the user, and the cognitive mastery probability value of 0.45 output from step 2.

[0080] Security Detection: The security detection module in the security and policy control unit is activated.

[0081] Keyword matching of the text: No preset high-risk words such as "short circuit", "smoke" or "high voltage contact" were found.

[0082] Hazard identification of images: No preset hazardous visual features such as sparks, abnormal smoke, or equipment overload were identified.

[0083] Judgment: The overall risk value does not exceed the preset safety threshold, the safety circuit breaker mechanism is not triggered, and the process continues.

[0084] Policy routing: The policy selector in the security and policy control unit routes the policy based on the probability value of 0.45 (skilled learner interval) and experimental step three, through a preset decision tree.

[0085] Decision Result: The system selects the skilled user heuristic template as the target strategy template. Prohibited paths are activated; the system explicitly prohibits the use of novice templates that directly provide answers or expert templates that discuss complex harmonic theories.

[0086] Step 4: Contextual retrieval and guided response generation.

[0087] Input: Experimental step three, heuristic template for experienced users.

[0088] Contextual Retrieval: Based on Experiment Step 3, the generative interaction unit retrieves the knowledge base context from the vector database. The key point is that the retrieval query is limited to information only related to Experiment Step 3 (measuring amplitude-frequency characteristics) and its preceding steps (Experiment Step 1: Circuit Construction, Experiment Step 2: Signal Source Setup). The system actively masks information from subsequent steps (Experiment Step 4: Data Analysis, Experiment Step 5: Report Writing) to prevent spoilers.

[0089] Prompt Construction and Generation: The system integrates the following elements to construct the Prompt sent to the Large Language Model (LLM):

[0090] Role instruction: "You are a lab tutor assistant."

[0091] Current scenario: "A student is performing the 'measuring amplitude-frequency response' step in an RLC series resonance experiment and observes a waveform with its top clipped."

[0092] Cognitive State and Strategy Instructions: "This student has a moderate grasp of this knowledge point. Please use heuristic questioning to guide him to discover the reasons himself, rather than giving him a direct conclusion."

[0093] Knowledge base context: Knowledge fragments retrieved from vector databases related to RLC circuits, oscilloscope measurements, signal amplitude limitations, etc.

[0094] Original user question: "Why is the top of the waveform flat?"

[0095] LLM Generation and Output: The large language model generates responses that conform to the strategy based on the structured Prompt above. For example: "Please observe the Vpp (peak-to-peak) value displayed on the oscilloscope. Is it close to or exceeds the supply voltage range of your signal source or amplifier circuit? Consider what might happen to the waveform when the signal amplitude exceeds the linear range that the circuit can handle?"

[0096] Step 5: Closed-loop feedback and cognitive state update.

[0097] User feedback: After thinking about it, the student replied: "Oh, I understand now. It's because the signal amplitude is too large, exceeding the op-amp's power supply voltage, which causes the output to be clipped."

[0098] Status Update: The system recognizes this feedback as "correct understanding." The dynamic cognitive modeling unit then triggers a BKT algorithm update.

[0099] The overall observation metric for this interaction (which integrates the correctness of text understanding and possible operational norms assessment) is set to a high value (e.g., 1.0) to indicate "mastery".

[0100] By combining the forgetting function F(Δt), the probability of the student's understanding of the knowledge point "clipping distortion" is updated from 0.45 to 0.55 according to the algorithm in step 2.

[0101] The updated probability value is written into the user's "cognitive state vector table" to guide the next interaction.

[0102] If, during the security check in step 3, the system detects a direct short circuit in the text or smoke-emitting features in the image, it will immediately trigger a circuit breaker. At this point, policy routing and LLM generation are completely bypassed, and the system directly outputs a preset hard-coded security warning message to the user.

[0103] By anchoring the physical state in step 1 and retrieval using context masking in step 4, the generation of the LLM is strictly limited to the knowledge of the current and completed experimental steps, fundamentally eliminating the possibility of premature or irrelevant answers due to information leakage.

[0104] The entire system forms a closed loop. Every interaction a student makes affects their cognitive state, which in turn directly determines the tutoring strategies for the next interaction, thus enabling a dynamic and personalized evolution of tutoring difficulty and methods.

[0105] One possible structure of the electronic device provided in this application includes: a processor, a memory, and a communication interface, which are interconnected and communicate with each other via a communication bus and / or other forms of connection mechanisms.

[0106] The memory includes one or more, which may be, but is not limited to, Random Access Memory (RAM), Read Only Memory (ROM), Programmable Read-Only Memory (PROM), Erasable Programmable Read-Only Memory (EPROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), etc. The processor and other possible components can access the memory, reading and / or writing data therein.

[0107] A processor may be one or more, and can be an integrated circuit chip with signal processing capabilities. The aforementioned processors can be general-purpose processors, including Central Processing Units (CPUs), Microcontroller Units (MCUs), Network Processors (NPs), or other conventional processors; they can also be special-purpose processors, including Neural-network Processing Units (NPUs), Graphics Processing Units (GPUs), Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. Furthermore, when there are multiple processors, some may be general-purpose processors, while others may be special-purpose processors.

[0108] A communication interface includes one or more devices that can be used to communicate directly or indirectly with other devices to exchange data. The communication interface may include interfaces for wired and / or wireless communication.

[0109] One or more computer program instructions may be stored in the memory, and the processor may read and execute these computer program instructions to implement the methods provided in the embodiments of this application.

[0110] Understandably, electronic devices can include more or fewer components, or different structures, and these components can be implemented using hardware, software, or a combination thereof. Electronic devices can be physical devices, such as PCs, laptops, tablets, mobile phones, servers, and embedded devices, or they can be virtual devices, such as virtual machines and virtualized containers. Furthermore, electronic devices are not limited to single devices; they can also be combinations of multiple devices or clusters of numerous devices.

[0111] This application provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of any of the methods described above.

[0112] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the displayed or discussed mutual couplings, direct couplings, or communication connections may be through some communication interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0113] Furthermore, 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 units can be selected to achieve the purpose of this embodiment according to actual needs.

[0114] Furthermore, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0115] In this document, relational terms such as first and second are used only to distinguish one entity or operation from another entity or operation, without necessarily requiring or implying any such actual relationship or order between these entities or operations.

[0116] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. An experimental process control and interactive guidance system, characterized in that, include: The multimodal intent parsing unit is used to receive and parse text and / or image input from the user, and, in conjunction with a preset experimental flowchart, determine the current physical experimental step in which the user is. The dynamic cognitive modeling unit, connected to the multimodal intent parsing unit, is used to run an improved Bayesian knowledge tracing algorithm based on the user's historical interaction data to calculate and dynamically update the probability value of the user's cognitive mastery of the knowledge points related to the current step. The security and policy control unit is connected to the multimodal intent parsing unit and the dynamic cognitive modeling unit respectively, and is used to: perform risk detection on user input based on a preset keyword library and / or image danger feature library; When the risk value exceeds a preset threshold, the safety circuit breaker mechanism is triggered, blocking the generation of a normal response; When the risk value is determined to be no more than a preset threshold, a target strategy template is selected from multiple preset tutoring strategy templates based on the numerical range of the cognitive mastery probability value output by the dynamic cognitive modeling unit. The generative interaction unit, connected to the security and policy control unit, is used to generate a guiding response for the user by calling a large language model based on the target policy template and the knowledge base context related to the current physical experiment step when the risk value is determined to be no more than a preset threshold.

2. The system as described in claim 1, characterized in that, The improved Bayesian knowledge tracing algorithm incorporates the user's rating results in the experimental operation dimension into the update calculation of the cognitive mastery probability value.

3. The system as described in claim 2, characterized in that, The improved Bayesian knowledge tracing algorithm also introduces a forgetting function to decay and correct the probability value of knowledge mastery based on time intervals.

4. The system as described in claim 1, characterized in that, The numerical range of the cognitive mastery probability value is used for strategy routing through a preset decision tree structure, wherein the numerical range is at least divided into a low probability range corresponding to novices, a medium probability range corresponding to proficient users, and a high probability range corresponding to expert users.

5. The system as described in claim 1, characterized in that, When the security circuit breaker mechanism is triggered, it directly outputs a preset hard-coded security warning message to the user.

6. The system as described in claim 1, characterized in that, The generative interaction unit is also used to retrieve the knowledge base context related to the current step and its predecessor steps from the vector database according to the current physical experiment step, and to shield the information of subsequent steps.

7. The system as described in claim 1, characterized in that, The multimodal intent parsing unit determines the current physical experiment step the user is in through image recognition technology and / or system background operation logs.

8. A method for experimental process control and interactive guidance, characterized in that, include: Receive and parse text and / or image input from the user, and combine it with a preset experimental flowchart to determine the current physical experimental step the user is in; Based on the user's historical interaction data, run the improved Bayesian knowledge tracing algorithm to calculate and dynamically update the probability value of the user's knowledge mastery of the relevant knowledge points in the current step; Risk detection is performed on user input based on a pre-set keyword library and / or image hazard feature library; When the risk value exceeds a preset threshold, the safety circuit breaker mechanism is triggered, blocking the generation of a normal response; When the risk value is determined to be no more than a preset threshold, a target strategy template is selected from multiple preset coaching strategy templates based on the numerical range of the cognitive mastery probability value. When the risk value is determined to be no more than a preset threshold, a large language model is invoked to generate a guiding response for the user based on the target strategy template and the knowledge base context related to the current physical experiment step.

9. An electronic device, characterized in that, include: A processor and a memory, the memory storing machine-readable instructions executable by the processor, which, when executed by the processor, perform the method as described in claim 8.

10. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method of claim 8.

Images