A socratic constraint reflection diagnostic risk planning method

Abstract

This invention discloses a diagnostic risk planning method based on Socratic constraint reflection, comprising: calculating the probability and uncertainty of a student's mastery of any skill point based on learning log data, and constructing a cognitive risk state; constructing a set of candidate teaching actions based on the cognitive risk state, and configuring cost, risk compression intensity, and similarity attributes between actions for each action; based on the set of candidate teaching actions, pre-setting a cognitive residual risk set objective function containing a risk compression principal term and a redundancy penalty term; calculating the marginal risk reduction of any teaching action based on the cognitive residual risk set objective function, and determining the comprehensive value by combining information gain score, stability verification score, and action cost; based on the comprehensive value, selecting teaching actions under budget constraints through an adaptive decreasing threshold mechanism to form a candidate action package; trimming the candidate action package to generate a set of mandatory teaching actions and a set of alternative teaching actions, and outputting a structured teaching plan.

Description

Technical Field

[0001] This invention relates to the field of online education technology, and in particular to a diagnostic risk planning method based on Socratic constraint reflection. Background Technology

[0002] In intelligent education scenarios, learning systems need to complete a closed-loop organization of "learning-practice-testing-evaluation-supplementation-review" within a limited time budget to support phased score improvement and process management. With the application of large-scale pre-trained language models in education, common system capabilities include: enhanced retrieval of explanations and Q&A, tool interfaces linked to question banks for assessment and grading, learning log analysis, Socratic interactive questioning, and consistency verification based on reflection and self-checking.

[0003] Among these, Socratic dialogues and reflective verification are high-information-density diagnostic actions that can improve the accuracy of identifying true mastery, but they also consume learning time and interactive resources, and the benefits fluctuate significantly depending on the student's state and the type of question. The core need arising from this is to select a set of action combinations from multiple candidate teaching actions within budget constraints, minimizing residual cognitive risk and prioritizing coverage of key weaknesses to ensure stable and controllable overall results.

[0004] Currently, existing large-scale model-based educational application solutions mainly suffer from the following defects and shortcomings: First, there is a lack of a mechanism for selecting diagnostic actions under budget constraints: existing systems typically employ fixed rules or localized triggering conditions when initiating diagnostic actions such as Socratic questioning or reflective verification. For example, they may uniformly question or reflect on every question, or only trigger these actions after a mistake is made. This type of strategy fails to comprehensively consider the time cost, interaction cost, and overall benefits of diagnostic actions. It cannot prioritize different learning projects under the constraint of a limited total time or number of attempts, which can easily lead to insufficient diagnosis of key weaknesses or excessive consumption of resources in non-critical areas.

[0005] Second, there is a lack of unified target modeling for residual cognitive risk: existing solutions mostly use accuracy, completion rate, or single test results as optimization targets, focusing on improving local indicators and lacking a holistic characterization of how much uncertainty or potential point loss risk still exists after completing a set of teaching actions. Due to the lack of a unified and optimizable objective function, the system struggles to make global trade-offs among various teaching actions and cannot determine from an overall perspective whether the current learning stage has reached an acceptable level of risk.

[0006] Third, there is a lack of robust mechanisms to handle fluctuations in diagnostic outcomes: The diagnostic effectiveness of Socratic questioning and reflective verification is significantly affected by various factors such as the student's immediate state, question type, language expression ability, contextual information, and the randomness of model generation. Existing systems typically make judgments based on the results of a single dialogue or reflection, lacking constraints and correction mechanisms for the instability of diagnostic results. This can easily lead to misjudgments of mastery, resulting in unnecessary repeated diagnoses or excessive consumption of learning time.

[0007] Fourth, there is a lack of controllable fallback strategies and explainable plan output structures: In real classrooms or intensive training camps, teachers and administrators typically need to clearly define explainable learning plans, including why specific follow-up questions, reflections, or exercises are scheduled for the current stage, and which tasks are mandatory and which can be supplemented when time is limited. However, existing systems mostly output results in the form of continuous dialogue chains or content recommendation lists, lacking tailorable and reviewable structured plans, making it difficult to meet the actual needs of teaching management and process supervision.

[0008] Therefore, there is an urgent need to propose a diagnostic risk planning method based on Socratic constraint reflection that is logically simple, accurate, and reliable. Summary of the Invention

[0009] To address the above problems, the present invention aims to provide a diagnostic risk planning method based on Socratic constraint reflection. The technical solution adopted by the present invention is as follows: A diagnostic risk planning method based on Socratic constraint reflection includes the following steps: Based on learning log data, calculate the probability and uncertainty of a student's mastery of any skill point, and construct a cognitive risk status. Based on the cognitive risk status, a set of candidate teaching actions is constructed, including exercises, micro-lessons, Socratic dialogues, and reflection verification. Each action is then configured with cost, risk compression intensity, and similarity attributes between actions. Based on the set of candidate teaching actions, a target function for the set of cognitive residual risks is preset, which includes risk compression main terms and redundant penalty terms; Based on the objective function of the cognitive residual risk set, the marginal risk reduction of any teaching action is calculated, and the comprehensive value is determined by combining the information gain score, stability verification score and action cost. Based on the comprehensive value, teaching actions are selected under budget constraints through an adaptive decreasing threshold mechanism to form a candidate action package; The candidate action package is trimmed to generate a set of mandatory teaching actions and a set of alternative teaching actions, and a structured teaching plan is output.

[0010] Compared with the prior art, the present invention has the following beneficial effects: This invention models Socratic dialogue and reflective verification as a diagnostic operator with higher cost but higher risk reduction efficiency. It constrains and schedules the triggering frequency of the operator by using the comprehensive value of unit cost, so that high-cost diagnostic actions are only selected when there is significant marginal benefit, avoiding abuse within the limited learning time and thus improving the overall learning efficiency.

[0011] This invention constructs a cognitive residual risk set objective function to uniformly characterize "the uncertainty and potential risk of losing points that still exist after completing a set of teaching actions," enabling learning plans to revolve around an optimizable overall goal. This facilitates global weighing and comparison among multiple types of teaching actions, unlike traditional solutions that rely on scattered indicators.

[0012] This invention introduces an exponentially decaying risk compression term into the objective function, reflecting the objective law that marginal returns decrease with increasing input. At the same time, through a redundancy penalty mechanism corresponding to action similarity, it suppresses the repeated selection of homogeneous teaching actions, so that the planning results have better stability and robustness under different student states and question types.

[0013] This invention adopts a dual-criteria output structure of "candidate action package - mandatory action set - alternative action set". While ensuring priority coverage of key weak points, it provides a clear and interpretable plan hierarchy for actual teaching execution. When there is insufficient available time or changes in the teaching rhythm, the mandatory action set provides a basic safety net to avoid the overall plan from failing.

[0014] In summary, this invention has the advantages of simple logic and high accuracy and reliability, and has high practical and promotional value in the field of online education technology. Attached Figure Description

[0015] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope of protection. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a logic flowchart of the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this application clearer, the present invention will be further described below with reference to the accompanying drawings and embodiments. The embodiments of the present invention include, but are not limited to, the following embodiments. All other embodiments obtained by those skilled in the art based on the embodiments in this application without inventive effort are within the scope of protection of this application.

[0018] like Figure 1 As shown, this embodiment provides a Socratic constrained reflection diagnostic risk planning method, targeting teaching scenarios with limited time and interaction resources. It models the learning process as an incompletely observable system where cognitive risk states evolve with teaching actions. Under budget constraints, it performs controlled scheduling of high-information-density diagnostic actions to achieve stable convergence of cognitive risk states. Specifically, it includes the following steps: The first step is to calculate the probability and uncertainty of a student's mastery of any skill point based on learning log data, and to construct a cognitive risk status.

[0019] In this step, a model is created based on the background log data generated by students during their actual learning process to model their current cognitive state for subsequent diagnostic action selection and planning decisions. The background logs include, but are not limited to: time spent answering questions, accuracy of answers, number of modifications, page or application switching behavior, copy-paste behavior, and usage prompts.

[0020] (11) Obtain skill points The corresponding cognitive risk quantity is expressed as follows: ; in, Students In time Next, regarding skill points The cognitive risk level is such that the higher the value, the more likely it is that points will be lost on that skill at the current stage, and the more uncertain the mastery is. Indicates skill points The weighting coefficient is used to reflect the importance of the skill point in the target test or teaching objective. The weighting source may include the distribution of test scores, frequency of occurrence, or the focus of teaching at different stages. Students In time Next, regarding skill points The probability of mastery is used to characterize the likelihood of correctly understanding and stably applying the skill point, and its value range is [0,1]. Students In time Next, regarding skill points The uncertainty is used to characterize the probability of correctly understanding and stably applying the skill point, and its value range is [0,1].

[0021] (12) Utilizing students In time Next, regarding skill points Cognitive risk level The total global cognitive risk can be calculated using the following expression: ; in, Students In time The total global cognitive risk is used to characterize the overall uncertainty and potential risk of losing points that still exist in the current learning stage; Represents a set of skill points.

[0022] The second step involves constructing a set of candidate teaching actions based on the perceived risk status, including exercises, micro-lessons, Socratic dialogues, and reflection verification, and configuring cost, risk compression intensity, and similarity attributes between actions for each action.

[0023] (21) Let the set of candidate teaching actions be . And any teaching action The teaching activities include practice exercises, micro-lessons, Socratic dialogues, and reflection and verification. Practice exercises include basic questions, variations, similar question groups, and subdivided challenging questions, used to test and reinforce specific skill points through answering. Micro-lessons include videos or segments explaining one or more skill points, used for clarifying knowledge, transferring methods, and supplementing problem-solving strategies. Socratic dialogues include a sequence of follow-up questions centered on specific weak skill points or key problem-solving steps, used to expose students' cognitive blind spots and hidden errors through continuous questioning. Reflection and verification include self-checking, restatement, consistency verification through rephrasing, and robustness verification of steps, used to determine whether students' current understanding is stable and to avoid situations where "it appears correct but lacks repeatability."

[0024] (22) Configure action cost, risk compression effect intensity and action similarity constraints for any teaching action.

[0025] For each teaching action in the candidate action set The system establishes the following attribute descriptions for it: First, action cost: defining teaching actions. The resource cost is This represents the amount of resources consumed in performing the action. These resources can be uniformly converted into learning time, interaction rounds, or computing power costs, and are used to impose overall budget constraints in subsequent steps.

[0026] Second, the intensity of risk compression effect: defining the action. skill points The intensity of the risk compression effect is , is a non-negative real number used to describe the execution of teaching actions. After that, regarding skill points The corresponding ability to reduce perceived risk. The higher the value, the more significant the effect of the action in reducing the risk of that skill point.

[0027] Third, action similarity constraint: To avoid repeatedly selecting highly similar teaching actions or diagnostic functions within a limited budget, an action similarity metric is introduced, defining any two actions... and The similarity between them is The value ranges from [0,1]. A larger value indicates that the two actions are more similar in terms of content, skill coverage, or diagnostic effect. This similarity can be calculated based on question features, skill point coverage, or semantic representation, and serves as a constraint factor to suppress the stacking of homogeneous actions in the subsequent action selection process.

[0028] The third step involves, based on the candidate teaching action set, pre-setting an objective function for the cognitive residual risk set, which includes a risk compression term and a redundancy penalty term. Here, the expression for the cognitive residual risk set objective function is: ; in, This indicates the execution of the current set of instructional actions. Post-cognitive residual risk value, , The smaller the value, the lower the risk that the current learning state has not yet been covered or confirmed; Indicates teaching actions skill points The intensity of the risk compression effect; This represents the exponential decay function, used to characterize the characteristic that the risk of the same skill point decreases with cumulative investment under the action of multiple related actions; This represents the redundancy penalty coefficient, used to adjust the intensity of the impact of action similarity on overall risk. Indicates teaching actions With teaching actions Similarity measurement between .

[0029] In the above definition, the first term of the objective function for the cognitive residual risk set is used to characterize the risk compression effect at each skill level: as teaching actions for the same skill point are continuously added, The first term gradually increases, causing the corresponding index term to decrease monotonically, reflecting the diminishing marginal returns characteristic during risk compression. The second term is used to discourage the repeated selection of highly similar teaching actions within a limited budget, avoiding resource waste caused by the stacking of homogeneous actions.

[0030] To ensure the effectiveness of subsequent planning and selection algorithms, the aforementioned cognitive residual risk set function... It has the following properties: 1) Nonnegativity That is, under any set of actions, the residual cognitive risk is not less than zero.

[0031] 2) The monotony and non-incrementality of the primary term, along with the set of actions. The extension, For any skill point Neither of these factors decreases, thus ensuring that the exponential decay term remains monotonically unchanged, guaranteeing that risk continues to decline as effective actions increase.

[0032] 3) Diversity constraint characteristics: By introducing a redundancy penalty term based on action similarity, the superposition effect of homogeneous actions is explicitly constrained in the overall objective function, thereby achieving a balance between risk reduction effect and action diversity, and avoiding the consumption of budget resources by low-value repetitive actions.

[0033] The fourth step involves calculating the marginal risk reduction of any instructional action based on the objective function of the cognitive residual risk set, and determining the comprehensive value by combining the information gain score, stability verification score, and action cost. This includes the following steps: (41) The marginal risk reduction of any teaching action is calculated, and its expression is: ; in, Indicates the current set of teaching actions Add teaching actions based on The larger the value of the reduction in residual cognitive risk after the action, the more significant the marginal contribution of the action to risk reduction at the current stage.

[0034] (42) To achieve a balance between risk reduction effect, diagnostic value, and resource cost, the method unifies the marginal risk reduction with the diagnostic value. The comprehensive value is determined by combining information gain score, stability verification score, and action cost, and its expression is: ; in, Indicates teaching actions In the current set of teaching actions Comprehensive value under certain conditions; Indicates teaching actions The efficiency of reducing risk per unit cost is used to reflect its cost-effectiveness under limited budget conditions; Indicates teaching actions The information gain score is used to characterize the potential ability of the action to expose key cognitive blind spots. The score can be estimated based on factors such as the degree of difference between the student's statement and factual evidence during the questioning process, and the coverage of key concepts. Indicates teaching actions The stability verification score is used to characterize the student's ability to confirm the stability of the action. The score can be estimated based on factors such as the consistency of the results of multiple restatements, the consistency of the conclusions drawn under different question formats, and the repeatability of the problem-solving steps. Indicates teaching actions Resource costs.

[0035] The fifth step involves using an adaptive decreasing threshold mechanism to filter teaching actions under budget constraints, forming candidate actions, based on the comprehensive value. In this step, to reduce the number of sequential decisions while ensuring risk reduction, this embodiment employs an adaptive selection mechanism combining threshold filtering and phased expansion. By gradually lowering the filtering threshold, actions meeting the conditions are selected in batches from the candidate teaching actions, thereby improving the parallelism and computational efficiency of the overall planning process. Specifically: (51) No teaching actions have been selected yet; the set of selected actions is denoted as an empty set. The preset initial screening threshold is... ;in, This indicates the teaching action when no teaching action is currently selected. The comprehensive value; This represents the set of empty actions.

[0036] (52) Gradually relax the screening criteria and introduce a stage index. Let the first The screening threshold for the stage is: Where E represents the tolerance error coefficient, which is used to control the threshold decrease rate, and its value ranges from (0,1). This is used to ensure that the threshold decreases monotonically as the stage progresses, thereby gradually incorporating teaching actions with lower overall value but still having risk reduction significance.

[0037] (53) Construct a candidate subset that passes the threshold condition: ;in, Indicates the first A set of candidate teaching actions that meet the requirement that their comprehensive value is not lower than the current screening threshold at each stage.

[0038] (54) Preset budget constraints: Where B represents the total budget allowed in this plan, which is used to limit learning time, number of interaction rounds, or equivalent resource consumption.

[0039] The sixth step involves trimming the candidate action packages to generate a set of mandatory teaching actions and a set of alternative teaching actions, outputting a structured teaching plan. In this embodiment, after completing the selection of teaching actions based on threshold filtering and budget constraints, to balance risk reduction and actual teaching feasibility, a dual-criteria result structure is introduced in the output stage. This structure divides the action selection results into candidate packages, a mandatory set, and an alternative set, thereby forming an interpretable and tailorable teaching plan output.

[0040] To improve the robustness of coverage of key risk points, this embodiment allows for a slight relaxation of budget constraints within a controlled scope, resulting in a candidate action package containing more potentially high-value actions, denoted as... In candidate actions Based on this, a set of essential teaching movements is obtained by trimming. and alternative teaching movements Its expression is: ; ; in, This indicates the candidate action package. Any subset of is used for evaluation as possible combinations of instructional actions during the cutting process; This represents the objective function for the set of cognitive residual risks defined in step three; This represents the cognitive residual risk function under budget constraints. The set of teaching actions that achieves the minimum value.

[0041] The above embodiments are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any changes made based on the design principles of the present invention, or any non-creative modifications made thereon, shall fall within the scope of protection of the present invention.

Claims

1. A diagnostic risk planning method based on Socratic constraint reflection, characterized in that, Includes the following steps: Based on learning log data, calculate the probability and uncertainty of a student's mastery of any skill point, and construct a cognitive risk status. Based on the cognitive risk status, a set of candidate teaching actions is constructed, including exercises, micro-lessons, Socratic dialogues, and reflection verification. Each action is then configured with cost, risk compression intensity, and similarity attributes between actions. Based on the set of candidate teaching actions, a target function for the set of cognitive residual risks is preset, which includes risk compression main terms and redundant penalty terms; Based on the objective function of the cognitive residual risk set, the marginal risk reduction of any teaching action is calculated, and the comprehensive value is determined by combining the information gain score, stability verification score and action cost. Based on the comprehensive value, teaching actions are selected under budget constraints through an adaptive decreasing threshold mechanism to form a candidate action package; The candidate action package is trimmed to generate a set of mandatory teaching actions and a set of alternative teaching actions, and a structured teaching plan is output.

2. The diagnostic risk planning method based on Socratic constraint reflection according to claim 1, characterized in that, Based on learning log data, the probability and uncertainty of a student's mastery of any skill point are calculated, and a cognitive risk status is constructed, including the following steps: Seek skill points The corresponding cognitive risk quantity is expressed as follows: ； in, Students In time Next, regarding skill points The amount of cognitive risk; Indicates skill points Weighting coefficients; Students In time Next, regarding skill points The probability of mastering; Students In time Next, regarding skill points Uncertainty; Using students In time Next, regarding skill points Cognitive risk level The total global cognitive risk can be calculated using the following expression: ； in, Students In time The total amount of overall perceived risk; Represents a set of skill points.

3. The diagnostic risk planning method based on Socratic constraint reflection according to claim 2, characterized in that, Based on the perceived risk status, a set of candidate teaching actions is constructed, including exercises, micro-lessons, Socratic dialogues, and reflection verification. Each action is then assigned cost, risk compression intensity, and similarity attributes between actions, including the following steps: Let the set of candidate teaching actions be And any teaching action The teaching actions include practice exercises, micro-lessons, Socratic dialogues, and reflection and verification. For any teaching action, constraints are set on action cost, risk compression effect intensity, and action similarity.

4. The diagnostic risk planning method based on Socratic constraint reflection according to claim 3, characterized in that, The expression for the objective function of the cognitive residual risk set is: ； in, This indicates the execution of the current set of instructional actions. Post-cognitive residual risk value, ; Indicates teaching actions skill points The intensity of the risk compression effect; Represents an exponentially decaying function; Indicates the redundancy penalty coefficient; Indicates teaching actions With teaching actions Similarity measurement between .

5. The diagnostic risk planning method based on Socratic constraint reflection according to claim 4, characterized in that, Based on the objective function of the cognitive residual risk set, the marginal risk reduction of any teaching action is calculated, and the comprehensive value is determined by combining the information gain score, stability verification score, and action cost, including the following steps: The marginal risk reduction of any teaching action can be expressed as follows: ； in, Indicates the current set of teaching actions Add teaching actions based on The resulting decrease in residual cognitive risk; The comprehensive value is determined by combining information gain score, stability verification score, and action cost, and its expression is as follows: ； in, Indicates teaching actions In the current set of teaching actions Comprehensive value under certain conditions; Indicates teaching actions The unit cost risk reduction efficiency; Indicates teaching actions Information gain score; Indicates teaching actions Stability verification score; Indicates teaching actions Resource costs.

6. The diagnostic risk planning method based on Socratic constraint reflection according to claim 5, characterized in that, Based on the comprehensive value, teaching actions are selected under budget constraints through an adaptive decreasing threshold mechanism to form a candidate action package, including the following steps: The preset initial screening threshold is ;in, This indicates the teaching action when no teaching action is currently selected. The comprehensive value; Represents the set of empty actions; Introducing a stage index Let the first The screening threshold for the stage is: Where E represents the tolerance error coefficient; Construct a candidate subset that meets the threshold conditions: ;in, Indicates the first A set of candidate teaching actions that meet the comprehensive value not lower than the current screening threshold at each stage; Preset budget constraints: Where B represents the total budget allowed for this plan.

7. The diagnostic risk planning method based on Socratic constraint reflection according to claim 6, characterized in that, The candidate action package is trimmed to generate a set of mandatory teaching actions and a set of alternative teaching actions, and a structured teaching plan is output, including the following steps: The candidate action package is marked as In candidate actions Based on this, a set of essential teaching movements is obtained by trimming. and alternative teaching movements Its expression is: ； ； in, Indicates candidate action package Any subset of.

Images