Scientific literacy assessment methods and electronic devices

Abstract

This application provides a scientific literacy assessment method and electronic device. The method includes: acquiring structured scientific theory data and unstructured scientific research practice data; constructing a scientific inquiry mind map based on the unstructured scientific research practice data, and comparing the scientific inquiry mind map with a standard map to obtain a topological evaluation score; determining whether a cognitive bias trigger signal is generated based on the structured scientific theory data and the topological evaluation score, and dynamically adjusting the fusion weights after the cognitive bias trigger signal is generated to determine the comprehensive literacy score; constructing a multidimensional cognitive bias vector for the target assessment sample that receives the cognitive bias trigger signal, and mapping the multidimensional cognitive bias vector to the corresponding scientific research defect label. This application can effectively improve the objectivity and accuracy of scientific literacy assessment.

Description

Technical Field

[0001] This application relates to the field of scientific literacy assessment technology, and more specifically, to a scientific literacy assessment method and electronic device. Background Technology

[0002] With the vigorous construction of science and technology middle school demonstration sites and the development of off-campus innovation training industries across the country, and with major universities relying on the "Strong Foundation Program" to carry out early selection of top innovative talents, scientific literacy assessment has become an important part of the routine teaching evaluation, selection of science and technology talents, and personalized training in primary and secondary schools.

[0003] Currently, the main methods for assessing students' scientific literacy include traditional comprehensive quality evaluation systems and learning analysis systems based on exam and homework data. Traditional comprehensive quality evaluation systems only input subject exam scores from the academic affairs system or rely on teachers' subjective scoring, resulting in a single data dimension and a relatively crude evaluation method. Learning analysis systems based on exam and homework data only analyze structured exercise data, limiting their application scenarios. These assessment methods lead to low objectivity and accuracy in scientific literacy assessment. Summary of the Invention

[0004] In view of this, the purpose of this application is to provide a scientific literacy assessment method and electronic device that can effectively improve the objectivity and accuracy of scientific literacy assessment.

[0005] In a first aspect, embodiments of this application provide a method for assessing scientific literacy, the method comprising: Acquire theoretical data from structural chemistry and practical data from unstructured scientific research; A scientific inquiry mind map is constructed based on the unstructured scientific research practice data, and the scientific inquiry mind map is compared with a standard map to obtain a topological evaluation score; Based on the structural chemistry subject theory data and the topology evaluation score, it is determined whether a cognitive bias trigger signal is generated, and the fusion weight is dynamically adjusted after the cognitive bias trigger signal is generated to determine the comprehensive literacy score. For the target evaluation sample that receives the cognitive bias trigger signal, a multidimensional cognitive bias vector is constructed, and the multidimensional cognitive bias vector is mapped to the corresponding scientific research defect label.

[0006] Secondly, embodiments of this application also provide a scientific literacy assessment device, the device comprising: The data acquisition module is used to acquire theoretical data in structural chemistry and practical data from unstructured scientific research. The data processing module is used to construct a scientific inquiry mind map based on the unstructured scientific research practice data, and compare the scientific inquiry mind map with the standard map to obtain a topological evaluation score; The score determination module is used to determine whether a cognitive bias trigger signal is generated based on the structural chemistry subject theory data and the topology evaluation score, and to dynamically adjust the fusion weight to determine the comprehensive literacy score after the cognitive bias trigger signal is generated. The tag generation module is used to construct a multidimensional cognitive bias vector for the target evaluation sample that has received the cognitive bias trigger signal, and to map the multidimensional cognitive bias vector to the corresponding scientific research defect tag.

[0007] Thirdly, embodiments of this application also provide an electronic device, including: a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor communicates with the memory via the bus. When the machine-readable instructions are executed by the processor, the steps of the scientific literacy assessment method described above are performed.

[0008] Fourthly, embodiments of this application also provide a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the steps of the scientific literacy assessment method described above.

[0009] The scientific literacy assessment method and electronic device provided in this application improve the objectivity and accuracy of scientific literacy assessment by integrating structured theoretical data and unstructured scientific research practice data, combining dynamic weighting mechanisms to suppress evaluation noise, and then achieving refined defect attribution through multidimensional cognitive bias vectors.

[0010] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0011] To more clearly illustrate the technical solutions of the embodiments of this application, 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 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.

[0012] Figure 1 One of the flowcharts for a scientific literacy assessment method provided in the embodiments of this application; Figure 2 A second flowchart of a scientific literacy assessment method provided in an embodiment of this application; Figure 3 A flowchart of a scientific literacy assessment method provided in the embodiments of this application is shown in part three. Figure 4 Flowchart four of a scientific literacy assessment method provided in the embodiments of this application; Figure 5 This is one of the structural schematic diagrams of a scientific literacy assessment device provided in the embodiments of this application; Figure 6 A second schematic diagram of a scientific literacy assessment device provided in an embodiment of this application; Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0013] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely represents selected embodiments of this application. Based on the embodiments of this application, every other embodiment obtained by those skilled in the art without inventive effort falls within the scope of protection of this application.

[0014] Please see Figure 1 , Figure 1 This is a flowchart illustrating a scientific literacy assessment method provided in an embodiment of this application. Figure 1 As shown in the embodiments of this application, the method includes: S101. Obtain theoretical data from structural chemistry and practical data from unstructured scientific research. S102. Construct a scientific inquiry mind map based on unstructured scientific research practice data, and compare the scientific inquiry mind map with the standard map to obtain the topological evaluation score; S103. Based on the theoretical data of structural chemistry and the topological evaluation score, determine whether a cognitive bias trigger signal is generated, and dynamically adjust the fusion weight after the cognitive bias trigger signal is generated to determine the comprehensive literacy score. S104. For the target evaluation sample that receives the cognitive bias trigger signal, construct a multidimensional cognitive bias vector and map the multidimensional cognitive bias vector to the corresponding scientific research defect label.

[0015] This application's embodiments effectively improve the objectivity and accuracy of scientific literacy assessment by integrating structured theoretical data with unstructured scientific research practice data, combining a dynamic weighting mechanism to suppress evaluation noise, and then using a multi-dimensional cognitive bias vector to achieve refined defect attribution.

[0016] The following is a detailed explanation of steps S101 to S104 above: In step S101, theoretical data of structural chemistry and practical data of unstructured scientific research are acquired.

[0017] In the above steps, the structural subject theory data refers to standardized and structured student subject learning data. This data has a unified format, clear dimensions, and can be directly quantified and statistically analyzed. It is primarily used to characterize students' level of theoretical knowledge mastery. In other words, structural subject theory data can comprehensively reflect students' theoretical knowledge reserves, providing a benchmark reference for subsequent cross-modal data fusion and bias judgment.

[0018] For example, the theoretical data of structural chemistry can include quantitative data such as students' standardized test scores at various stages, mid-term and final exams, subject competition scores, knowledge mastery rates, and question type scores. The data can come from the school's academic management system and official learning assessment platform. After standardized system entry and compliance verification, there is no human subjective tampering or subjective scoring bias, and the data has extremely high credibility.

[0019] In addition, the theoretical data of structural chemistry can be flexibly adapted to the data dimensions according to the evaluation scenario. For example, for the selection scenario of top talents in the university's Strong Foundation Program, high-end structured and quantitative data such as discipline independent enrollment test scores and discipline competition level scores can be added to further adapt to the evaluation needs of high-level talents and ensure the comprehensiveness and adaptability of theoretical data in different scenarios.

[0020] In the above steps, unstructured scientific research practice data refers to the text-based practical products generated by students during scientific inquiry, innovation experiments, and scientific research practice. These products lack a fixed format and are not standardized, making them difficult to quantify and statistically analyze directly using traditional mathematical models. They are primarily used to characterize students' practical inquiry logic and scientific thinking processes, and are key data for measuring students' scientific inquiry abilities. This type of data truthfully records students' complete experimental inquiry process and thinking process, and can intuitively reflect the rationality and rigor of students' scientific research logic, thus contributing to the goal of refined evaluation.

[0021] For example, unstructured scientific research practice data mainly includes pure text content such as experimental inquiry logs, scientific and technological innovation project practice records, experimental reports, and research notes written by students themselves. The text format is free, the expression is colloquial, and the content has no uniform standard. It may have personalized characteristics such as jumps in thinking, incomplete process records, and unclear variable descriptions, and cannot be directly quantitatively analyzed.

[0022] Optionally, unstructured scientific research practice data can be adapted to various scientific and technological innovation practice scenarios, such as covering regular classroom experiments, after-school scientific and technological innovation exploration, specialized scientific research training in off-campus training institutions, and research projects by top talents, which can comprehensively capture students' scientific research thinking performance in different scenarios. For example, in distinctive scientific and technological innovation projects carried out by off-campus training institutions, such as robot programming exploration, ecological environment surveys, and new material exploration, the text content recorded by students on the practice process, problem analysis, and improvement ideas can all be used as unstructured scientific research practice data input into the system for evaluation.

[0023] In step S102, a scientific inquiry mind map is constructed based on unstructured scientific research practice data, and the scientific inquiry mind map is compared with the standard map to obtain a topological evaluation score.

[0024] In the above steps, the Scientific Inquiry Thinking Flow Graph (SITFG) refers to a directed acyclic topological structure graph representing the logical thinking of students' scientific inquiry, generated by the element extraction model. With standardized inquiry nodes at its core and temporally logically connected edges as links, it can completely reconstruct the students' experimental inquiry process and thought path, thereby transforming unstructured text into quantifiable topological data. This Scientific Inquiry Thinking Flow Graph can intuitively and structurally present the complete logical chain of students' scientific research inquiry.

[0025] Optionally, a scientific inquiry mind map can fully replicate the entire thought process of students, from observing phenomena, proposing hypotheses, conducting practical operations, and summarizing and revising. Each node and connecting edge corresponds to the student's actual inquiry behavior and thought logic. Optionally, a scientific inquiry mind map can possess the characteristics of standardization, comparability, and quantification, enabling topological comparison with standard inquiry maps preset by industry experts. By quantifying the rationality and rigor of students' inquiry logic through topological differences, it can be achieved.

[0026] Furthermore, the standard mind map can be a standardized topological map jointly constructed by scientific research and education experts and renowned science and technology innovation teachers, conforming to the rigorous logic of scientific inquiry. It covers all categories of experimental inquiry scenarios in primary and secondary schools and science and technology innovation education, serving as a benchmark reference system for evaluating the standardization and rigor of students' scientific research inquiry logic. The standard mind map can solidify the optimal inquiry sequence, node composition, and logical relationships of various scientific experiments, providing a unified and objective evaluation standard for the scoring and deviation determination of students' mind maps, thereby avoiding the bias of subjective human scoring.

[0027] For example, standard topological maps can be categorized and constructed according to disciplines and experiment types, covering routine inquiry experiments and scientific innovation experiments in multiple disciplines such as physics, biology, chemistry, and geography. Each experimental scenario can correspond to a set of standardized topological maps, at least one standardized topological map, or a set of standardized maps maintained by discipline, experiment type, grade level, and scenario version. These maps clearly define the essential inquiry nodes, temporal sequence, and logical relationships of the experiment. In addition, standard maps can be dynamically optimized and upgraded according to updates to the teaching syllabus and iterations of scientific innovation inquiry scenarios, continuously adapting to the latest science education requirements and inquiry paradigms, ensuring the timeliness and professionalism of the evaluation criteria.

[0028] In the above steps, comparing the scientific inquiry mind map with the standard diagram yields a topological evaluation score. The topological evaluation score is a quantitative score used to objectively evaluate the rigor, completeness, and rationality of the student's scientific inquiry logic, obtained by quantitatively calculating the differences in topological structure between the student's generated scientific inquiry mind map and the standard diagram. The score range can be customized; a higher score indicates that the student's inquiry logic closely aligns with the standard scientific research paradigm and demonstrates stronger rigor in thinking. This topological evaluation score enables the calculability of unstructured scientific research practices.

[0029] For example, topology evaluation scores can be calculated using a graph editing distance algorithm. By statistically analyzing topological differences between student mind maps and standard graphs, such as missing nodes, redundant nodes, temporal discrepancies, and logical mismatches, the degree of difference is quantified and converted into corresponding scores, achieving a completely objective and subjectively unbiased scoring system. Furthermore, topology evaluation scores can be precisely calibrated using core parameters such as node classification thresholds and topology similarity calculation weights. Differentiated weights are assigned to inquiry nodes of varying importance, with higher deduction weights for defects in core logical nodes and lower deduction weights for deviations in secondary process nodes, ensuring that the scoring results align with actual teaching needs. For instance, the absence of core inquiry nodes such as control variables and control experiment settings will significantly reduce the topology score, while redundant node deviations such as experimental notes and descriptions of secondary phenomena have minimal impact on the score.

[0030] In one alternative embodiment, such as Figure 2 As shown, step S102 specifically includes: Step S1021: For unstructured scientific research practice data, use the element extraction model to extract the research nodes and construct a scientific research mind map; Step S1022: Perform topological similarity calculation between the scientific inquiry mind map and the standard map to obtain the topological evaluation score.

[0031] In step S1021, a scientific inquiry mind map can be constructed using an element extraction model. This element extraction model is a specialized element extraction model built upon a multimodal large language model, adapted to the logic of scientific inquiry. Through preset Prompt engineering constraints and scientific inquiry rules, it can accurately extract elements, classify nodes, and logically organize unstructured scientific research texts. It possesses a unique ability to recognize scientific inquiry logic, effectively solving the problems of general models lacking understanding of scientific logic and exhibiting low extraction accuracy. In other words, the element extraction model can accurately break down messy practical texts and organize standardized inquiry process nodes.

[0032] Specifically, the element extraction model can be based on mainstream large language models and fine-tuned through customized scientific inquiry Prompt instructions to accurately identify inquiry actions, thought processes, and temporal logic in scientific research texts. For example, for photosynthesis experiment logs written by students, the element extraction model can accurately remove irrelevant colloquial expressions and extract core inquiry elements such as experimental hypotheses, experimental operations, experimental phenomena, and conclusion revisions, while filtering out invalid and redundant information.

[0033] Optionally, the element extraction model can be equipped with a dedicated scientific inquiry element recognition lexicon and logical judgment rules to adapt the extraction logic to different disciplines and types of scientific experimental inquiry scenarios. This effectively improves the accuracy and adaptability of element extraction in different scenarios, thereby avoiding the problems of poor scenario adaptability, element omission, and incorrect extraction of general models.

[0034] Among these methods, the element extraction model is used to extract inquiry nodes. Inquiry nodes refer to the core thinking and operational units extracted from unstructured scientific research practice texts that constitute a complete scientific inquiry process. They are the basic components for constructing mind maps and can accurately break down students' fragmented and disjointed inquiry thinking. Inquiry nodes eliminate colloquial, redundant, and ineffective expressions in the text, retaining only standardized and core inquiry elements.

[0035] Optionally, the exploration nodes can be extracted according to the general logic of scientific inquiry, covering the core links of the entire experimental process and ensuring the integrity of the thought process breakdown. For example, in the mechanics experiment log written by students, the system can accurately extract various core exploration nodes such as observation of experimental phenomena, assumption of force laws, operation of experimental equipment, data recording and summarization, and conclusion correction.

[0036] In one optional embodiment, step S1021 includes the following steps: The element extraction model is driven by preset constraint instructions to extract elements from unstructured scientific research practice data, resulting in research elements. These elements are then categorized and grouped according to preset node classification criteria. Groups of research elements that do not meet the judgment conditions are eliminated based on preset node classification thresholds. The remaining research element groups are then encapsulated to form standardized research nodes. Standardized research nodes include observation nodes, hypothesis nodes, operational verification nodes, and inductive correction nodes. The sequential logic and temporal dependencies of each standardized research node in the scientific research practice process are analyzed to obtain the logical association mapping relationship between nodes. Based on the logical association mapping relationship, directional connection edges are established between corresponding nodes according to the experimental research process. All standardized research nodes and directional connection edges together generate a scientific research mind map with a directed acyclic structure.

[0037] Among them, the preset constraint instructions refer to Prompt engineering instructions specifically customized for the extraction of scientific inquiry elements. These instructions can include standardized constraints such as the definition of scientific inquiry, the scope of core elements, extraction rules, and elimination conditions. They are used to standardize the extraction behavior of large models, avoiding the chaotic and illogical content extracted by general models, and ensuring the professionalism and accuracy of element extraction. These preset constraint instructions are the core constraints that adapt to scientific inquiry scenarios and achieve accurate element extraction. Specifically, the preset constraint instructions can explicitly limit the extraction to elements corresponding to the four core stages of scientific inquiry, filtering out redundant content such as colloquial comments, ineffective sentiments, and irrelevant descriptions. Furthermore, the preset constraint instructions can be refined and iterated according to subject-specific scenarios, such as configuring exclusive constraint rules for the inquiry characteristics of different disciplines like physics, chemistry, and biology, further improving the accuracy of element extraction.

[0038] Here, the research elements are the original core content units of scientific research that have been initially extracted by the model but have not been classified or organized. They form the basis for standardized research nodes and can include various core research information such as experimental phenomena, hypotheses, operational steps, data conclusions, and revised ideas. They have not been classified or screened in compliance with regulations.

[0039] Specifically, the elements of inquiry can be derived from complete unstructured experimental texts, covering all the core behaviors and thinking content of student inquiry, ensuring the integrity of subsequent node construction. For example, the hypotheses "light affects the photosynthetic rate of plants," the operation of "planting plants under different light and watering conditions," and the phenomenon "sun-loving plants grow faster" extracted from students' photosynthesis logs are all original elements of inquiry.

[0040] Furthermore, the pre-defined node classification criteria refer to a system of classification rules developed based on the general paradigm of scientific inquiry to distinguish the four standardized inquiry nodes. These criteria clearly define the scope, content characteristics, and core attributes of each type of node, providing a unified standard for the accurate classification of inquiry elements. This system aligns with the formal scientific experimental inquiry process, ensuring the professionalism and standardization of node classification.

[0041] Here, the pre-defined node classification criteria can clearly define the exclusive standards for the four major nodes: the observation node corresponds to the observation of phenomena and data recording before and during the experiment; the hypothesis node corresponds to the experimental conjecture and research hypothesis proposed based on existing knowledge; the operation verification node corresponds to the specific experimental operation, variable control, control setting, and data collection behavior; and the summary and correction node corresponds to the data analysis, conclusion summary, hypothesis verification, and idea correction after the experiment.

[0042] Furthermore, pre-defined node classification criteria can distinguish the affiliation of similar elements, avoid node classification confusion, and ensure classification accuracy. For example, "pre-defined experimental hypothesis" and "final experimental conclusion" are respectively classified into the hypothesis node and the inductive correction node, eliminating element misclassification.

[0043] The preset node classification threshold refers to a quantitative threshold used to determine the validity of research elements and screen qualified nodes. It is used to eliminate invalid research elements that are incomplete, vaguely described, or lack scientific research value, ensuring the validity and standardization of standardized nodes. The preset node classification threshold can be flexibly adjusted according to the required evaluation accuracy. For example, the preset node classification threshold can include an element completeness threshold and a content validity threshold to directly eliminate incomplete or vague elements.

[0044] Here, the remaining research elements are packaged into standardized research nodes. Standardized research nodes refer to standardized core nodes that conform to the general scientific research paradigm after being classified, screened and packaged. They can include four fixed types: observation nodes, hypothesis nodes, operation verification nodes, and inductive correction nodes.

[0045] In other words, the four types of standardized inquiry nodes can fully cover the entire scientific inquiry process; none can be omitted, and together they constitute a complete logical chain of scientific research. For example, a complete photosynthesis experiment node includes: observation node (differences in plant growth status under light), hypothesis node (the stronger the light, the faster the photosynthesis), operational verification node (planting plants under different light and watering conditions), and inductive correction node (verifying the hypothesis based on the growth results). The system can accurately identify the standardized nodes that students are missing, providing a direct basis for subsequent defect diagnosis. For example, if a student lacks the inductive correction node, it can be directly determined that there is an inquiry defect of "missing experimental conclusion summary."

[0046] Furthermore, by analyzing the sequential logic and temporal dependencies of each standardized research node in the scientific research process, we obtain the logical association mapping relationship between each node. The logical association mapping relationship here refers to the inherent temporal sequence, causal deduction, and logical dependency relationship between each standardized research node. It is the core basis for building the directional connection of the mind map and fits the progressive logic of thinking in real experimental research.

[0047] Optionally, the logical mapping relationships can follow a fixed scientific research paradigm: observation of phenomena → hypothesis formulation → experimental verification → inductive correction, forming a progressive logical chain without logical inversion, jumps, or gaps. For example, phenomenon observation and hypothesis formulation must be completed first, followed by experimental operations, and finally, conclusions must be summarized. The system can accurately identify abnormal logical mapping relationships, such as operations preceding hypotheses or conclusions drawn directly without observation, providing a basis for topology scoring and defect determination.

[0048] Furthermore, directional connecting edges are established between corresponding nodes according to the experimental inquiry process. Directional connecting edges are node connecting lines with directional indicators built according to temporal and logical relationships. They are used to represent the progressive logic and sequence between nodes. All standardized inquiry nodes and directional connecting edges together form a directed acyclic graph structure, thus obtaining a scientific inquiry mind map.

[0049] Here, directional connecting edges can be unidirectionally connected according to the exploration sequence, pointing from the preceding node to the following node, ensuring the chronological rigor of the scientific inquiry mind map. For example, a hypothetical node can point to an operation verification node via a directional connecting edge, and the operation verification node can point to an inductive correction node. Furthermore, missing, disordered, and redundant directional connecting edges can be accurately identified by the system as core differences for topological similarity calculation.

[0050] In this embodiment, firstly, a preset constraint instruction drives the element extraction model to extract elements from unstructured scientific research practice data, filtering out all inquiry elements. Then, based on preset node classification criteria, all inquiry elements are categorized into four groups: observation nodes, hypothesis nodes, operational verification nodes, and inductive correction nodes. Combined with preset node classification thresholds, vaguely described or invalid inquiry elements are eliminated, while valid inquiry elements are retained and encapsulated into four standardized inquiry nodes. Next, the temporal logic of each inquiry node is analyzed to determine the positive logical association mapping relationship of "observation-hypothesis-operation-induction". Finally, based on this logical association mapping relationship, unidirectional directional connection edges are built between adjacent inquiry nodes, ultimately generating a structurally complete and logically clear directed acyclic scientific inquiry mind map, accurately reproducing the student's inquiry thinking and operational process.

[0051] In step S1022, the scientific inquiry mind map and the standard graph are subjected to topological similarity calculation. Topological similarity calculation refers to the calculation process of quantifying the overall difference between the scientific inquiry mind map and the expert standard graph in terms of node composition, number of nodes, temporal order, logical connection edges, etc., through professional graph structure comparison algorithm, so as to accurately judge the standardization of the inquiry logic from the topological structure level.

[0052] For example, topological similarity calculation can employ a graph edit distance algorithm. This algorithm quantifies overall similarity by statistically analyzing differences such as missing nodes, redundant nodes, temporal discrepancies, and logical edge mismatches. The more differences, the lower the similarity and the lower the topological score. Furthermore, topological similarity calculation can be configured with differentiated weighting parameters. High weights can be assigned to structural differences in core exploration nodes, while low weights can be assigned to differences in secondary nodes, ensuring that the scoring results align with the teaching focus.

[0053] In step S103, based on the structural chemistry subject theory data and topological evaluation scores, it is determined whether a cognitive bias trigger signal is generated, and the fusion weight is dynamically adjusted after the cognitive bias trigger signal is generated to determine the comprehensive literacy score.

[0054] The cognitive bias trigger signal is an anomaly signal automatically generated by the system when it detects a serious mismatch between students' theoretical knowledge and scientific research practice abilities, exhibiting abnormal characteristics of high scores but low practical skills. This signal marks the presence of evaluation noise in the target assessment sample, necessitating weight correction and score calibration. This cognitive bias trigger signal is the core identifier distinguishing normal assessment samples from abnormal biased samples, providing a triggering basis for subsequent dynamic weight adjustments and refined bias diagnosis. It effectively solves the problem of traditional assessments failing to identify the disconnect between theoretical and practical abilities.

[0055] Optionally, the generation of cognitive bias trigger signals can be based on the differential matching relationship between theoretical data and practical topology scores. When a student's structural chemistry subject performance is at the upper level of the group, but their topology evaluation score is significantly lower than the passing threshold, indicating a serious gap between theoretical and practical abilities, the system can immediately generate a cognitive bias trigger signal to mark the sample as having a high-score, low-ability anomaly. Optionally, the cognitive bias trigger signal can be adapted to the judgment criteria of different assessment scenarios. For different scenarios such as the selection of top talents, routine academic assessment, and screening of scientific and technological innovation capabilities, the theoretical score ranking threshold and the topology score passing threshold can be dynamically adjusted to ensure the accurate adaptability of anomaly judgment.

[0056] In the above steps, for anomalous samples with cognitive biases, a preset algorithm is used to nonlinearly adjust the fusion weights of structural chemistry subject theory data and topology evaluation scores, reducing the weight proportion of inflated theoretical scores and thus suppressing evaluation noise caused by high scores but low abilities. This weight correction mechanism can break the limitations of traditional fixed-weight evaluation, realize dynamic adaptation and fusion of cross-modal data, and significantly improve the authenticity of the evaluation.

[0057] For example, based on the baseline fusion weights output by the Bayesian network, the theoretical data weights can be penalized and reduced using a nonlinear exponential decay function. The more severe the deviation, the greater the weight reduction, ensuring that the final fusion score more closely reflects the student's actual scientific research ability, thus eliminating evaluation errors. Specifically, the adjustment magnitude of the dynamically corrected fusion weights is positively correlated with the posterior probability of cognitive bias. The system can achieve fine-grained gradient weight adjustment based on the severity of the deviation, rather than a fixed magnitude adjustment, ensuring the rationality and accuracy of the score correction. For instance, samples with slight theoretical-practice deviations only have their theoretical weights slightly reduced, while samples with severe deviations have their theoretical weights significantly reduced, achieving differentiated and precise noise suppression.

[0058] Among them, the comprehensive literacy score is the final quantitative score that can truly reflect the comprehensive level of students' scientific literacy after cross-modal data fusion and bias weight correction. This score has the characteristics of high objectivity and high authenticity, taking into account students' theoretical knowledge reserves and scientific research and practical inquiry capabilities, and can solve the problem of distorted results caused by the single-dimensional scoring of traditional assessments.

[0059] For example, the comprehensive quality score can be calculated using a baseline weight fusion method for normal samples and a corrected weight fusion method for biased samples, adapting to the evaluation needs of different samples and ensuring the authenticity of the scores of all samples.

[0060] In one alternative embodiment, such as Figure 3 As shown, step S103 specifically includes the following steps: Step S1031: Input the structural chemistry discipline theory data and topology evaluation scores into the pre-constructed Bayesian network dynamic weight model, and output the cognitive bias posterior probability and benchmark fusion weight.

[0061] In the above steps, the Bayesian network dynamic weight model is a dedicated model built on Bayesian probability theory for cross-modal data fusion and bias probability determination. By presetting prior probability parameters and combining the matching relationship between theoretical data and practical topology scores, it dynamically calculates the posterior probability of cognitive bias and outputs a benchmark fusion weight that is adapted to the current sample.

[0062] Optionally, the Bayesian network dynamic weight model can incorporate prior probability parameters specific to scientific literacy assessment. Trained and optimized based on massive amounts of teacher and student assessment data, it can accurately adapt to the data characteristics of science and technology innovation education scenarios, avoiding the problem of poor adaptability of general models. For example, the model can be trained on a large amount of sample data showing "high theoretical scores but low practical scores," accurately identifying the bias probability of such abnormal samples.

[0063] Step S1032: When the posterior probability of cognitive bias is greater than the preset judgment threshold, a cognitive bias trigger signal for the corresponding target evaluation sample is generated, and the fusion weight corresponding to the structural science theoretical data is corrected by an exponential decay function based on the benchmark fusion weight.

[0064] In the above steps, the posterior probability of cognitive bias is the true probability that a student has a cognitive bias between theoretical and practical abilities, calculated by the Bayesian network model based on the input cross-modal data. Here, a higher posterior probability of cognitive bias indicates a greater likelihood that the student has high scores but low practical abilities and flawed inquiry logic.

[0065] Optionally, the posterior probability of cognitive bias can range from 0 to 1, where 0 represents no bias and 1 represents the absolute presence of bias. The system can quantify the degree of bias using precise probability values. For example, if a student's posterior probability of bias is 0.92, it means that there is a 92% probability that they have a high-scoring but low-ability cognitive bias, classifying them as a severely biased sample.

[0066] Among them, the baseline fusion weight is a standardized, fixed-fit cross-modal data fusion weight output by the Bayesian network model for normal samples without cognitive bias. It includes the weight of structural disciplinary theory data and the weight of topological evaluation scores, which is suitable for the assessment scenarios of most normal students.

[0067] Optionally, the benchmark fusion weights, trained and optimized with a massive amount of normal samples, can balance the assessment of theoretical knowledge and practical inquiry abilities. For example, in a typical scenario, the benchmark weights can be set to 40% for theoretical scores and 60% for topology practice scores, emphasizing the assessment of students' scientific research and practical inquiry abilities. Furthermore, the benchmark fusion weights can be flexibly adjusted according to the assessment scenario; for elite talent selection scenarios, the weight of the practical topology score can be increased, while for routine learning assessments, the weights of the two can be appropriately balanced.

[0068] Here, the preset judgment threshold refers to the probability critical value used to determine whether cognitive bias exists. It is the core boundary distinguishing normal samples from abnormal biased samples. The preset judgment threshold can be flexibly configured according to the evaluation accuracy and scenario requirements. For example, in a conventional evaluation scenario, the preset judgment threshold can be set to 0.7. When the posterior probability of the bias is greater than 0.7, a significant cognitive bias is determined to exist, and a cognitive bias trigger signal is generated; if it is lower than 0.7, it is determined to be a normal sample. For high-precision, top-notch selection scenarios, the preset judgment threshold can be increased to strictly screen high-quality talents, avoid the selection of pseudo-high-scoring talents, and improve the accuracy of selection.

[0069] In the above steps, the exponential decay function refers to a non-linear penalty function used for correcting the weights of outlier samples. Using the baseline weight as a base, it achieves gradient decay of the theoretical data weights according to the magnitude of the deviation probability. The more severe the deviation, the greater the weight decay, accurately suppressing high-scoring but low-ability noise. Here, the exponential decay function is constructed based on confidence levels; the higher the posterior probability of the deviation, the greater the theoretical score weight decay coefficient, and the lower the final theoretical weight, effectively weakening the interference of inflated theoretical scores on the overall score. The exponential decay function enables refined gradient weight adjustment, avoiding the score distortion problem caused by fixed-amplitude weight adjustment, and ensuring that the corrected score closely reflects the student's true ability.

[0070] Step S1033: The structural chemistry subject theory data and topological evaluation scores are weighted and fused using the modified fusion weights to obtain the comprehensive quality score.

[0071] For example, the system inputs a student's structural chemistry score of 95 and topology evaluation score of 30 into a Bayesian network dynamic weight model. Based on built-in prior probability parameters, the model calculates a posterior probability of 0.94 for the student's cognitive bias and outputs the baseline fusion weights for a typical scenario (40% theory, 60% practice). The system determines that the posterior probability of 0.94 is greater than a preset threshold of 0.7 and immediately generates a cognitive bias trigger signal. Subsequently, it calls an exponential decay function to penalize the theoretical data weights using the baseline weights as a base, significantly reducing the theoretical score weights and moderately increasing the topology practice score weights. Finally, the corrected weights are used to perform a weighted fusion calculation on the two sets of data, resulting in a true comprehensive competency score of 65, thus completing the score calibration for the abnormal sample.

[0072] The embodiments of this application also include: when the posterior probability of cognitive bias is not greater than a preset judgment threshold, the benchmark fusion weight is directly used to perform weighted fusion calculation on the structural chemistry subject theory data and the topological evaluation score to obtain the comprehensive literacy score.

[0073] Here, for normal student samples with a high degree of matching between theoretical knowledge and scientific research inquiry ability, the posterior probability of cognitive bias calculated by the system using a Bayesian network model is less than or equal to a preset judgment threshold. This indicates that such students do not exhibit high-scoring but low-ability cognitive bias, and their inquiry logic matches well with their theoretical knowledge reserves, thus requiring no weight penalty correction. The system directly calls the baseline fusion weights output by the model and, according to the preset theoretical and practical score proportions, performs a weighted summation calculation on the structural chemistry subject theory data and topological evaluation scores to obtain an objective and accurate comprehensive quality score.

[0074] For example, a student scored 88 points in biology theory and 85 points in the topological evaluation of photosynthesis experiments. The posterior probability of the bias was only 0.21, which did not reach the preset judgment threshold. The system directly used the benchmark fusion weight fusion calculation to finally obtain a comprehensive literacy score of 86.8 points, which accurately reflects the student's balanced scientific literacy level.

[0075] The above method enables precise scoring of normal and abnormal samples, ensuring the objectivity and authenticity of the evaluation results for the entire sample.

[0076] In step S104, for the target evaluation sample that receives the cognitive bias trigger signal, a multidimensional cognitive bias vector is constructed, and the multidimensional cognitive bias vector is mapped to the corresponding scientific research defect label.

[0077] In the above steps, the target assessment sample refers to student assessment samples that the system determines have cognitive biases or have triggered abnormal signals. These samples exhibit problems such as a disconnect between theoretical and practical abilities and flawed research logic, making them the core targets for refined attribution diagnosis and targeted teaching intervention. By accurately selecting the target assessment sample, resources can be precisely focused, ineffective calculations on normal samples can be avoided, and overall assessment and diagnosis efficiency can be improved.

[0078] Here, only samples that successfully trigger abnormal signals will proceed to the subsequent deviation vector construction and defect diagnosis process. Normally matched samples do not require refined attribution; only a comprehensive competency score is output. Optionally, target evaluation samples can be batch-screened and processed, supporting learning assessments at the class, grade, and school levels. This allows for the rapid identification of student groups with research logic deficiencies, providing data support for optimizing group-based science and technology innovation teaching. For example, the system can batch-screen evaluation data for an entire grade level, uniformly identifying all high-scoring but low-ability target samples with research logic deficiencies, and generating a group defect statistical report.

[0079] The multidimensional cognitive bias vector refers to a multidimensional feature vector that integrates horizontal group comparison characteristics and vertical historical trend characteristics. It can centrally represent the degree of abnormal deviation and the trend of change in students' current scientific research ability. This multidimensional cognitive bias vector breaks through the limitations of traditional single static assessment, realizing the combination of static bias and dynamic trend, and providing data support for refined and comprehensive attribution of deficiencies.

[0080] Optionally, the feature dimensions of the multidimensional cognitive bias vector can be flexibly expanded according to the evaluation accuracy requirements. Sub-features such as experiment completion, node completeness, and logical error rate can be added to further improve the refinement of bias characterization.

[0081] Among them, the research deficiency label is obtained through model mapping and precise classification. It is used to concretely represent the specific logical defects in students' scientific research. It has uniqueness, accuracy, and intervention capability, and can solve the technical defects of traditional assessments that only output scores and cannot attribute causes. Each research deficiency label corresponds to a clear research question and teaching improvement direction, providing a precise basis for subsequent targeted teaching intervention and closed-loop verification.

[0082] Optionally, the research defect tags cover common logical problems across all categories of scientific and technological innovation research in primary and secondary schools, including standardized defect types such as multivariate confusion, failure to control variables, lack of awareness of control groups, inverted experimental logic, data fabrication, lack of practical procedures, and one-sided conclusion summarization. The tags are clearly defined and the problems are clearly identified. Furthermore, the research defect tags can be continuously expanded according to the scenarios of scientific and technological innovation research, adapting to the complex defect identification needs of advanced scientific and technological innovation experiments and innovative research projects, and continuously enriching the defect tag system.

[0083] In one alternative embodiment, such as Figure 4 As shown, step S104 specifically includes the following steps: Step S1041: Select target evaluation samples with cognitive abnormalities based on cognitive bias trigger signals.

[0084] Step S1042: Calculate the KL divergence based on the current evaluation data of the target evaluation sample and the population norm data to obtain the lateral deviation feature, and fit the slope of the score change based on the historical evaluation data of the target evaluation sample to obtain the longitudinal trend feature.

[0085] Among them, the group norm data refers to the standardized assessment statistics of all students in the corresponding grade, the corresponding academic stage, and the corresponding assessment scenario. It includes statistical indicators such as the mean, variance, distribution interval, and node completion rate of the group topology score. It serves as the benchmark reference data for judging whether the ability of an individual student deviates from the group level.

[0086] Specifically, the population norm data is dynamically updated in real time, automatically iterating and updating after each batch evaluation to ensure the timeliness and accuracy of the reference standards. For example, after the entire grade completes the photosynthesis experiment evaluation, the system automatically calculates the topology score distribution and node accuracy rate of all students in the grade, generating the population norm data for this experiment. In addition, the population norm data can be further subdivided by grade level and scenario, distinguishing between different grade levels such as primary school, junior high school, and senior high school, as well as different scenarios such as regular teaching, science and technology innovation training, and elite selection, to ensure reference adaptability.

[0087] In the above steps, KL divergence is a statistical indicator used to quantify the difference between the distribution of an individual student's current assessment data and the group's norm data. The larger the KL divergence value, the more serious the deviation of the student's inquiry ability and data characteristics from the group average level, and the higher the degree of anomaly. It is a core quantitative parameter for characterizing lateral ability deviation. Optionally, KL divergence can accurately quantify the differences between students' topological node completion rate, logical error rate, and score level and the group, comprehensively reflecting lateral deviation characteristics.

[0088] Among them, lateral deviation features refer to static deviation features obtained through KL divergence quantification, which characterize the difference between a student's current single-investigation ability and the average level of the peer group, and can accurately reflect the degree of abnormality in a student's ability within the group. Optionally, lateral deviation features focus on static anomalies in a single assessment, accurately locating the individualized deficiencies in a student's current experimental investigation that distinguish them from the majority of students.

[0089] Among them, the longitudinal trend feature refers to the slope of change obtained by fitting students' scores in multiple historical assessments, which represents the dynamic trend of changes in students' scientific research ability. A positive slope indicates that the ability is steadily improving, a negative slope indicates that the ability is continuously declining, and a slope close to 0 indicates that the ability is stagnant.

[0090] For example, the longitudinal trend feature is generated by fitting data from students' recent standardized science and technology innovation assessments. The data sample is sufficient, the trend fit is accurate, and it can truly reflect the dynamic changes in students' abilities. For instance, if a student's topology score has been declining continuously in the last 5 experiments, and the fitting slope is negative, the longitudinal trend feature indicates an abnormal decline in ability.

[0091] Step S1043: Combine and merge the lateral deviation features with the vertical trend features to generate a multidimensional cognitive bias vector.

[0092] Step S1044: Use a hybrid attribution model to map and classify the multidimensional cognitive bias vector to obtain the corresponding scientific research defect labels.

[0093] Among them, the hybrid attribution model refers to a composite defect diagnosis model that integrates a rule engine and a machine learning classifier. It combines human-made scientific research rules with the classification capabilities of intelligent algorithms to achieve high-precision and highly interpretable defect label mapping, balancing the rigor of rules with the accuracy of algorithms. Here, the hybrid attribution model can avoid the shortcomings of a single rule engine (poor flexibility) and a single classifier (weak interpretability), achieving complementary advantages and significantly improving the accuracy and reliability of defect diagnosis.

[0094] In this embodiment, the system first filters out high-scoring but low-ability target evaluation samples through cognitive bias trigger signals. It then retrieves the student's evaluation data from the current photosynthesis experiment and the norm data for the entire grade level to calculate the KL divergence value, generating the student's current lateral group deviation characteristics. Simultaneously, it retrieves the student's recent topology evaluation scores from multiple scientific innovation experiments, obtains the slope of score changes through linear fitting, and generates longitudinal ability trend characteristics. The system concatenates the lateral deviation characteristics and longitudinal trend characteristics dimensionally, fusing them to generate a complete multidimensional cognitive bias vector to comprehensively characterize the student's static biases and dynamic ability anomalies. Finally, this multidimensional cognitive bias vector is input into a hybrid attribution model. Through rule verification and algorithm classification by the hybrid attribution model, the corresponding multivariate confounding research defect labels are accurately mapped, completing refined attribution diagnosis.

[0095] Specifically, hybrid attribution models include rule engines and classifiers.

[0096] Here, the rule engine pre-inputs prohibited behaviors, essential operational requirements, and basic logical norms for the entire scientific experiment process. It then performs item-by-item verification on the nodes, operational compliance, and data rationality represented by the multidimensional cognitive bias vector. Abnormal vector data caused by non-subjective inquiry issues such as abnormal data collection, incomplete log content, and system transmission failures are removed, and only valid vectors that can truly reflect the deficiencies in students' inquiry abilities are retained.

[0097] The classifier is pre-trained using a massive dataset of labeled multidimensional cognitive bias vectors and corresponding research defect labels. This covers various refined defect types, including multivariate confusion, logical inversion, lack of control group awareness, and biased conclusions, enabling it to accurately capture subtle feature differences within the vectors. For example, for composite defect vectors exhibiting both improper variable control and biased experimental conclusions, the rule engine cannot determine the defect using a single rule. The classifier, however, can identify primary and secondary defects through feature analysis and complete label matching.

[0098] Based on the above, step S1044 specifically includes: The rule engine is used to filter out violations and anomalies in the multidimensional cognitive bias vector to obtain compliant target multidimensional cognitive bias vectors. For target multidimensional cognitive bias vectors that meet the preset defect judgment conditions, the rule engine directly generates corresponding scientific research defect labels. For target multidimensional cognitive bias vectors that do not meet the preset defect judgment conditions, the target multidimensional cognitive bias vector is input into a classifier. The classifier calculates the matching probability of each scientific research defect label and selects the label with the highest matching probability as the final scientific research defect label of the target evaluation sample.

[0099] In the above steps, the rule engine, based on its built-in judgment rules, performs legality and validity checks on the original multidimensional cognitive bias vectors, distinguishing between genuine inquiry-related deficiencies and non-subjective abnormal data, and removing invalid interference data. This setup avoids interference from erroneous data generated during data collection, transmission, and parsing, ensuring that the vector data entering the defect judgment stage truly and objectively reflects the problems students encounter in their scientific inquiry process, thus improving the accuracy of defect label identification from the data source. The filtering process does not tamper with the valid feature values ​​within the vectors; it only removes invalid dimensions, corrects obvious data errors, and marks and isolates abnormal samples, ensuring the integrity of the original bias features.

[0100] Furthermore, a compliant target multidimensional cognitive bias vector is obtained through violation and anomaly filtering. This compliant target multidimensional cognitive bias vector possesses complete dimensions, valid values, and reliable features, making it a standardized feature vector that truly represents students' scientific inquiry biases and ability deficiencies. This target multidimensional cognitive bias vector retains all valid horizontal deviation features and vertical trend features of the original vector, eliminating external interference factors. It serves as the core input data for subsequent defect determination, and all defect label generation and classification operations are based on this compliant vector. The compliant vector strictly adheres to preset data format and numerical range requirements, with unified dimensionality, feature meaning, and numerical range, ensuring standardization of rule determination and classifier operations.

[0101] The preset defect judgment criteria are quantitative feature thresholds and logical judgment standards developed by domain experts based on experimental standards and teaching experience, targeting typical and explicit scientific inquiry defects. These criteria are integrated into the rule engine to quickly identify high-identifiability, high-occurrence hard inquiry questions. The judgment criteria use the feature values ​​and dimensional combinations of a multidimensional cognitive bias vector as the basis for judgment. As long as the compliant target multidimensional cognitive bias vector meets the corresponding threshold or logical relationship, the defect type can be directly identified without further classifier computation, balancing judgment efficiency and result intuitiveness. For example, the preset defect judgment criteria can be divided into multiple categories according to defect type, such as data fabrication, lack of practical process, complete logical inversion, and missing control group, with each category having independent judgment criteria.

[0102] In the above steps, the classifier calculates the probability value of a sample belonging to a certain research defect label based on the input compliant target multidimensional cognitive bias vector features. The value ranges from 0 to 1; the closer the value is to 1, the higher the probability that the sample corresponds to that type of defect, and the closer the value is to 0, the lower the correlation. Here, the correlation degree of different defect types is quantified by matching probability, achieving objective and quantifiable label selection and avoiding bias caused by subjective human judgment. Specifically, the matching probability is calculated independently for each type of research defect label, without interference, and can clearly reflect the tendency of the sample's defects. Finally, the label corresponding to the highest matching probability is selected as the final research defect label for the target evaluation sample.

[0103] In one optional embodiment, the embodiments of this application can match corresponding intervention strategies based on scientific research defect labels and generate isomorphic variant verification experiments.

[0104] Intervention strategies refer to targeted teaching improvement plans, tutoring approaches, and training directions that are pre-linked and stored in a knowledge graph, corresponding one-to-one with various research deficiency tags. They serve as the core guiding principle for personalized, individualized instruction, accurately matching students' specific inquiry deficiencies and addressing the problems of traditional, general, and untargeted teaching interventions. Here, intervention strategies can align with the principles of subject-based creative teaching in primary and secondary schools, adapt to students' cognitive levels, and directly provide precise guidance for teachers' lesson preparation and students' independent training.

[0105] Specifically, intervention strategies can be tailored to specific improvement plans for different research defect labels, with each defect corresponding to a unique and precise intervention approach, avoiding generic and template-based teaching guidance. For example, for the defect label "DefectMultiVariable_Confusion" (failure to control variables), the intervention strategy using knowledge graph matching is to strengthen training in single-variable control experiments, master the logic of setting up control experiments, and clearly identify the unique and irrelevant variables in the experiment.

[0106] Optionally, the intervention strategy can include tiered guidance, adapting the intervention intensity to students of different grades and ability levels. Lower grades focus on basic practical logic training, while higher grades focus on training in complex variable analysis and experimental design optimization. For example, to address the multivariate confusion deficit in elementary school, the intervention strategy emphasizes practical training in simple single-variable experiments; to address the same deficit in high school, the intervention strategy adds training in variable decomposition, control, and analysis of complex multivariate experiments.

[0107] Here, isomorphic variation verification experiments refer to novel, miniaturized experimental questions generated by fine-tuning the experimental scenario, parameters, and questioning methods, based on the core inquiry logic of the original student defect experiment. These questions retain the core inquiry knowledge points and experimental logical framework, and are used to specifically verify the effectiveness of students' defect rectification, achieving closed-loop verification. These experiments share the same logic as the original experiment but feature variations in the scenario, enabling precise assessment of students' mastery of the defective knowledge points and avoiding ineffective repetitive training.

[0108] Specifically, isomorphic variation verification experiments can rigorously address the core issues of deficiencies and precisely target the loopholes in students' inquiry logic, making them highly targeted. For example, in response to the deficiencies in students' photosynthesis experiments, such as confusion of multiple variables and failure to control the water variable, the system generates isomorphic variation verification exercises: "If you want to verify the effect of light, please set the watering amount for plant B in the dark room (fill in the blank: ___ ml) if plant A is watered with 100 ml of water." This tests the logic of controlling a single variable and corresponds perfectly to the deficiencies of the original experiment.

[0109] Here, isomorphic variation verification experiments can adopt a miniaturized and lightweight design, with simple and targeted question types. No complex experimental equipment is required, allowing students to quickly complete answers and verifications, adapting to daily learning progress screening and personalized training scenarios. For example, variation exercises generated to address control variable failure defects can be completed online through fill-in-the-blank and selection-type questions, eliminating the need for offline experiments and significantly improving the efficiency of closed-loop training.

[0110] Specifically, the above steps involve retrieving intervention strategies that match the research deficiency label from a pre-constructed knowledge graph of intervention strategies, generating an intervention report based on the matched intervention strategies, and, in conjunction with the parameters of the original experimental scenario, calling the experimental generation model to generate isomorphic variant verification experiments and accompanying exercises corresponding to the original experiment, under the constraints of preset generation rules and experimental templates.

[0111] The intervention strategy-related knowledge graph is a pre-built domain knowledge graph designed for science literacy teaching intervention scenarios. Using research defect tags as the core entity, it links various entities and relationships, including defect causes, teaching intervention approaches, training points, guidance techniques, and typical cases, forming an integrated "phenomenon-cause-strategy" knowledge network. This knowledge graph can integrate the guidance experience of renowned science and technology innovation teachers, the teaching requirements of the new curriculum standards, and experimental error correction training programs, achieving precise association between research defect tags and intervention strategies. It serves as the core data carrier for automated targeted intervention. The entity relationships within the knowledge graph are standardized and structured, supporting rapid retrieval, linked navigation, and content expansion. It can cover common defects in scientific inquiry across all subjects in primary and secondary schools and corresponding intervention plans, while also supporting the dynamic addition of defect tags and supporting strategies, demonstrating excellent scalability.

[0112] Optionally, the intervention strategy association knowledge graph can use each standardized scientific research defect label as an independent index node. Each label node can be associated with branches such as the cause of the defect, core intervention strategies, classroom tutoring suggestions, self-training directions, and reminders of common mistakes, forming a complete knowledge link.

[0113] In the above steps, the system uses research defect tags as search keywords to match corresponding entities and extract related content in the intervention strategy association knowledge graph. Relying on the indexing mechanism of the intervention strategy association knowledge graph, it achieves rapid location and can simultaneously retrieve the entire chain of content, including defect causes, intervention strategies, and guidance suggestions. The search results are complete and highly relevant, providing all the original materials for the generation of the intervention report.

[0114] Specifically, the intervention report is a standardized graphic report automatically generated based on intervention strategies, causes of deficiencies, and problem analysis retrieved from a knowledge graph. It is tailored to three user groups—teachers, students, and parents—with distinct content focuses, clearly explaining students' inquiry deficiencies, the root causes of problems, directions for improvement, and learning suggestions. The intervention report is highly interpretable, moving beyond simple score displays. It allows for the explanation of professional deficiency labels in plain language and provides actionable learning guidance, serving as a direct link between diagnostic results and teaching interventions.

[0115] Furthermore, by combining the parameters of the original experimental scenario and under the constraints of preset generation rules and experimental templates, the experimental generation model is invoked to generate isomorphic variant verification experiments and accompanying exercises corresponding to the original experiment.

[0116] Here, the original experimental scenario parameters can include structured parameters such as experiment name, experimental discipline, experimental equipment, research topic, variable settings, experimental conditions, and response format. Retaining the original experimental scenario parameters ensures that the variant experiment maintains consistency with the core research logic and scenario of the original experiment, preventing the variant experiment from deviating from the original's key assessment points and thus ensuring that the verification training is targeted. Taking the plant photosynthesis experiment as an example, the original experimental scenario parameters could include: Experiment name: Investigation of plant photosynthesis; Research topic: The effect of light on photosynthetic rate; Original variables: Light and water; Experiment format: Offline written log recording, etc.

[0117] The preset generation rules and experimental templates constitute a pre-defined constraint system for experiment generation, consisting of two parts: rule clauses and standardized templates. The generation rules define the difficulty range, question types, knowledge scope, variable modification rules, and answering requirements for variant experiments. The experimental templates are general frameworks designed for different disciplines and experiment types, including fixed formats such as question structure, question format, answer area, and score settings. The combination of these two constraints on the output of the experiment generation model ensures that isomorphic variant experiments have standardized question types, appropriate difficulty, and accurate test points, avoiding questions that are out of syllabus, off-topic, or unusual. For example, the preset generation rules explicitly require isomorphic variant verification experiments to retain the core knowledge points explored in the original experiment, only adjusting the scenario, parameters, and question format, without changing the core logical test points. Simultaneously, the question types are limited to lightweight question types such as fill-in-the-blank, multiple choice, and short answer, adapting to online rapid-response scenarios.

[0118] Here, the experiment generation model is a proprietary content generation model derived from a large language model combined with fine-tuning of the Prompt project. Under multiple constraints of generation rules, experiment templates, and original experiment scenario parameters, it automatically creates isomorphic variation verification experiments and accompanying exercises. This model is deeply adapted to the logic of scientific experiment question setting, strictly focusing on the knowledge points corresponding to scientific research deficiencies, ensuring a high degree of matching between exercises and deficiencies in diagnosis results. At the same time, it can generate exercises in batches and flexibly adjust the difficulty.

[0119] Furthermore, while maintaining the core inquiry logic of the original experiment, adjustments were made to the experimental parameters, scenario descriptions, and question formats to create new, lightweight experimental questions. This resulted in isomorphic variation verification experiments and accompanying exercises, used to verify whether students had understood and corrected any flaws in the inquiry logic. These exercises share the same logic as the original experiment but differ in format, belonging to the targeted verification question type. Specifically, the isomorphic variation verification experiments primarily consist of miniaturized written exercises, requiring no physical experimental equipment. Students can quickly complete their answers online, and the system automatically grades and compiles the results, completing a closed-loop verification process.

[0120] For example, the following is a complete illustration of the overall process using a plant photosynthesis research experiment scenario: The system first collects structured theoretical data and unstructured research practice data from a student. The student's biology theory score is 95, placing them in the top 20% of their grade. The unstructured data consists of a photosynthesis experiment log written by the student: "Hypothesis: The stronger the light, the faster the photosynthesis. I put plant A in the sun and plant B in a dark room. I watered A with 100ml of water and plant B with 50ml of water. As a result, plant A grew faster, proving the hypothesis is correct." The system then uses a feature extraction model to process the log, extracting four core nodes: observation, hypothesis, operational verification, and inductive correction. A corresponding scientific inquiry mind map is constructed, and its topological similarity is calculated against the standard photosynthesis experiment map. Due to the logical flaw of the student's experimental operation nodes involving simultaneous changes to both light and water variables, the topological structure differs significantly from the standard map, ultimately resulting in a topological evaluation score of 30 / 100. Next, the system inputs the student's 95-point subject theory score and 30-point topology evaluation score into a Bayesian network dynamic weight model. This identifies a significant disconnect between the student's theoretical and practical abilities, triggering a cognitive bias signal. Through an exponentially decaying nonlinear weight correction mechanism, the system penalizes the student's subject score by reducing its weight, ultimately correcting the initially inflated comprehensive literacy score to 65 points, effectively suppressing evaluation noise caused by high scores but low abilities. The system then marks this student as the target evaluation sample, retrieving evaluation data from the photosynthesis experiment of all students in the grade as the group norm to calculate the student's horizontal KL divergence deviation feature. Simultaneously, it retrieves the student's evaluation scores from the last five scientific innovation experiments, fitting a longitudinal slope feature of ability change. These two features are then concatenated to generate a multidimensional cognitive bias vector unique to the student. Finally, the system uses a hybrid attribution model to analyze and classify the bias vector, accurately matching the research deficiency labels indicating failed control variables, thus completing the entire process of scientific literacy assessment, bias calibration, and refined attribution. After obtaining the research defect tags indicating failed control variables from students, the system retrieves a pre-constructed knowledge graph of intervention strategies, accurately matches the corresponding intervention strategy, and generates an intervention report containing defect cause analysis, improvement ideas, and training priorities. The report clearly points out that the student's core problem is "failure to control the water variable to be unique during the experiment, while simultaneously modifying both light and water variables, resulting in the experimental results failing to verify the preset hypothesis, indicating a logical flaw in the control variables." Subsequently, the system, combining the scenario parameters of the original photosynthesis experiment and under the constraints of the preset experiment generation template and rules, calls the experiment generation model to generate the aforementioned isomorphic variation verification experiment exercises, which are then accurately pushed to the student for completion.If a student correctly fills in 100ml, it proves that they have mastered the single-variable control logic, and the defect rectification verification is passed; if the answer is wrong, the system will continue to push similar variant training until the student fully masters the core inquiry logic, and finally form a complete teaching closed loop of "assessment-diagnosis-intervention-verification" to solve the technical problems of existing technologies that only assess but do not teach and lack intervention.

[0121] This application's embodiments, by integrating structured theoretical data with unstructured scientific research practice data, combining a dynamic weighting mechanism to suppress evaluation noise, and then achieving refined defect attribution through a multi-dimensional cognitive bias vector, effectively improve the objectivity and accuracy of scientific literacy assessment. It solves the shortcomings of traditional assessments, such as fragmented heterogeneous data and misjudgments of high scores but low abilities, enhances the refined attribution capability of the assessment, abandons the traditional extensive scoring model, and achieves a technological leap from single score output to specific defect localization. Simultaneously, it addresses the core issues of existing technologies' inability to quantify the exploration logic of unstructured scientific research texts and lack of refined defect diagnosis capabilities, providing precise data support for subsequent personalized teaching interventions.

[0122] Based on the same inventive concept, this application also provides a scientific literacy assessment device corresponding to the scientific literacy assessment method. Since the principle of the device in this application is similar to the scientific literacy assessment method described above in this application, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be described again.

[0123] Please see Figure 5 , Figure 5 This is a schematic diagram of the structure of a scientific literacy assessment device provided in an embodiment of this application. Figure 5 As shown, the device 500 includes: Data acquisition module 501 is used to acquire theoretical data in structural chemistry and practical data in unstructured scientific research. Data processing module 502 is used to construct a scientific inquiry mind map based on the unstructured scientific research practice data, and compare the scientific inquiry mind map with a standard map to obtain a topological evaluation score; The score determination module 503 is used to determine whether a cognitive bias trigger signal is generated based on the structural chemistry subject theory data and the topology evaluation score, and to dynamically adjust the fusion weight to determine the comprehensive literacy score after the cognitive bias trigger signal is generated. The tag generation module 504 is used to construct a multidimensional cognitive bias vector for the target evaluation sample that has received the cognitive bias trigger signal, and to map the multidimensional cognitive bias vector to the corresponding scientific research defect tag.

[0124] Furthermore, such as Figure 6 As shown, the device 500 further includes an experimental verification module 505, which is used for: Based on the research defect labels, corresponding intervention strategies are matched, and isomorphic variant verification experiments are generated.

[0125] In one optional embodiment, the data processing module 502 is specifically used for: For the unstructured scientific research practice data, the element extraction model is used to extract the research nodes and construct a scientific research mind map; The topological similarity calculation is performed between the scientific inquiry mind map and the standard map to obtain the topological evaluation score.

[0126] In one optional embodiment, the data processing module 502 is further configured to: The unstructured scientific research practice data is processed by a feature extraction model driven by preset constraint instructions to obtain the research features. The research elements are categorized and grouped according to preset node classification criteria. Research element groups that do not meet the judgment conditions are eliminated by combining preset node classification thresholds. The remaining research element groups are packaged into standardized research nodes. The standardized research nodes include observation nodes, hypothesis nodes, operation verification nodes, and inductive correction nodes. The logical sequence and temporal dependency of each standardized research node in the scientific research practice process are analyzed to obtain the logical association mapping relationship between each node; Based on the logical association mapping relationship, directional connection edges are established between corresponding nodes according to the experimental investigation process. All standardized investigation nodes and directional connection edges are used to generate a scientific investigation mind map with a directed acyclic structure.

[0127] In one optional embodiment, the score determination module 503 is specifically used for: The structural chemistry theory data and the topology evaluation score are input into a pre-constructed Bayesian network dynamic weight model, which outputs the posterior probability of cognitive bias and the baseline fusion weight. When the posterior probability of the cognitive bias is greater than the preset judgment threshold, a cognitive bias trigger signal for the corresponding target evaluation sample is generated, and the fusion weight corresponding to the structural science theoretical data is corrected by an exponential decay function based on the benchmark fusion weight. The structural chemistry subject theory data and the topology evaluation score are weighted and fused using the modified fusion weights to obtain the comprehensive literacy score.

[0128] In an optional embodiment, the score determination module 503 is further configured to: When the posterior probability of the cognitive bias is not greater than the preset judgment threshold, the benchmark fusion weight is directly used to perform weighted fusion calculation on the structural chemistry subject theory data and the topology evaluation score to obtain the comprehensive literacy score.

[0129] In one optional embodiment, the label generation module 504 is used to: Based on the cognitive bias triggering signal, target evaluation samples with cognitive abnormalities are obtained by screening. The KL divergence is calculated based on the current evaluation data of the target evaluation sample and the population norm data to obtain the lateral deviation feature, and the slope of the score change is fitted based on the historical evaluation data of the target evaluation sample to obtain the longitudinal trend feature. The lateral deviation feature and the vertical trend feature are concatenated and fused to generate a multidimensional cognitive bias vector; The multidimensional cognitive bias vector is mapped and classified using a hybrid attribution model to obtain the corresponding scientific research defect labels.

[0130] In an optional embodiment, the hybrid attribution model includes a rule engine and a classifier, and the label generation module 504 is further configured to: The rule engine is used to perform violation and anomaly filtering on the multidimensional cognitive bias vector to obtain a compliant target multidimensional cognitive bias vector. For a target multidimensional cognitive bias vector that meets the preset defect judgment conditions, the rule engine directly generates a corresponding scientific research defect label. For a target multidimensional cognitive bias vector that does not meet the preset defect judgment conditions, the target multidimensional cognitive bias vector is input into the classifier. The classifier is used to calculate the matching probability of each scientific research defect label. The label corresponding to the highest matching probability is selected as the final scientific research defect label of the target evaluation sample.

[0131] In one optional embodiment, the experimental verification module 505 is specifically used for: Retrieve intervention strategies that match the research deficiency tags from a pre-built knowledge graph of intervention strategies, and generate an intervention report based on the matched intervention strategies; Based on the parameters of the original experimental scenario, and under the constraints of preset generation rules and experimental templates, the experimental generation model is invoked to generate isomorphic variant verification experiments and accompanying exercises corresponding to the original experiment.

[0132] The device provided in this application integrates structured theoretical data with unstructured scientific research practice data, combines a dynamic weighting mechanism to suppress evaluation noise, and then achieves refined defect attribution through a multi-dimensional cognitive bias vector. This effectively improves the objectivity and accuracy of scientific literacy assessment, solves the defects of traditional assessments such as fragmented heterogeneous data and misjudgment of high scores but low abilities, enhances the refined attribution capability of assessment, abandons the traditional extensive scoring mode, and achieves a technological leap from single score output to specific defect location. At the same time, it solves the core problems of existing technologies that cannot quantify the exploration logic of unstructured scientific research texts and lack refined defect diagnosis capabilities, providing precise data support for subsequent personalized teaching intervention.

[0133] Please see Figure 7 , Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 7 As shown, the electronic device 700 includes a processor 710, a memory 720, and a bus 730.

[0134] The memory 720 stores machine-readable instructions executable by the processor 710. When the electronic device 700 is running, the processor 710 communicates with the memory 720 via the bus 730. When the machine-readable instructions are executed by the processor 710, they can perform the operations described above. Figures 1 to 4 The steps of the scientific literacy assessment method in the illustrated method embodiment can be found in the method embodiment for specific implementation methods, which will not be repeated here.

[0135] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, can perform the above-described actions. Figures 1 to 4 The steps of the scientific literacy assessment method in the illustrated method embodiment can be found in the method embodiment for specific implementation methods, which will not be repeated here.

[0136] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0137] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, 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 shown 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.

[0138] 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.

[0139] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0140] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a processor-executable, non-volatile, 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 portion 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.

[0141] Finally, it should be noted that the above-described embodiments are merely specific implementations of this application, used to illustrate the technical solutions of this application, and not to limit them. The scope of protection of this application is not limited thereto. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features, within the scope of the technology disclosed in this application. Such modifications, changes, 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, and should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for assessing scientific literacy, characterized in that, The method includes: Acquire theoretical data from structural chemistry and practical data from unstructured scientific research; A scientific inquiry mind map is constructed based on the unstructured scientific research practice data, and the scientific inquiry mind map is compared with a standard map to obtain a topological evaluation score; Based on the structural chemistry subject theory data and the topology evaluation score, it is determined whether a cognitive bias trigger signal is generated, and the fusion weight is dynamically adjusted after the cognitive bias trigger signal is generated to determine the comprehensive literacy score. For the target evaluation sample that receives the cognitive bias trigger signal, a multidimensional cognitive bias vector is constructed, and the multidimensional cognitive bias vector is mapped to the corresponding scientific research defect label.

2. The method according to claim 1, characterized in that, The method further includes: Based on the research defect labels, corresponding intervention strategies are matched, and isomorphic variant verification experiments are generated.

3. The method according to claim 1, characterized in that, The process of constructing a scientific inquiry mind map based on the unstructured scientific research practice data and comparing the scientific inquiry mind map with a standard map to obtain a topological evaluation score includes: For the unstructured scientific research practice data, the element extraction model is used to extract the research nodes and construct a scientific research mind map; The topological similarity calculation is performed between the scientific inquiry mind map and the standard map to obtain the topological evaluation score.

4. The method according to claim 3, characterized in that, The process of extracting inquiry nodes and constructing a scientific inquiry mind map using an element extraction model for the unstructured scientific research practice data includes: The unstructured scientific research practice data is processed by a feature extraction model driven by preset constraint instructions to obtain the research features. The research elements are categorized and grouped according to preset node classification criteria. Research element groups that do not meet the judgment conditions are eliminated by combining preset node classification thresholds. The remaining research element groups are packaged into standardized research nodes. The standardized research nodes include observation nodes, hypothesis nodes, operation verification nodes, and inductive correction nodes. The logical sequence and temporal dependency of each standardized research node in the scientific research practice process are analyzed to obtain the logical association mapping relationship between each node; Based on the logical association mapping relationship, directional connection edges are established between corresponding nodes according to the experimental investigation process. All standardized investigation nodes and directional connection edges are used to generate a scientific investigation mind map with a directed acyclic structure.

5. The method according to claim 1, characterized in that, The step of determining whether a cognitive bias trigger signal is generated based on the structural chemistry subject theory data and the topology evaluation score, and dynamically adjusting the fusion weights to determine the comprehensive literacy score after the cognitive bias trigger signal is generated, includes: The structural chemistry theory data and the topology evaluation score are input into a pre-constructed Bayesian network dynamic weight model, which outputs the posterior probability of cognitive bias and the baseline fusion weight. When the posterior probability of the cognitive bias is greater than the preset judgment threshold, a cognitive bias trigger signal for the corresponding target evaluation sample is generated, and the fusion weight corresponding to the structural science theoretical data is corrected by an exponential decay function based on the benchmark fusion weight. The structural chemistry subject theory data and the topology evaluation score are weighted and fused using the modified fusion weights to obtain the comprehensive literacy score.

6. The method according to claim 5, characterized in that, The method further includes: When the posterior probability of the cognitive bias is not greater than the preset judgment threshold, the benchmark fusion weight is directly used to perform weighted fusion calculation on the structural chemistry subject theory data and the topology evaluation score to obtain the comprehensive literacy score.

7. The method according to claim 1, characterized in that, For the target evaluation sample that receives the cognitive bias trigger signal, a multidimensional cognitive bias vector is constructed, and the multidimensional cognitive bias vector is mapped to the corresponding scientific research defect label, including: Based on the cognitive bias triggering signal, target evaluation samples with cognitive abnormalities are obtained by screening. The KL divergence is calculated based on the current evaluation data of the target evaluation sample and the population norm data to obtain the lateral deviation feature, and the slope of the score change is fitted based on the historical evaluation data of the target evaluation sample to obtain the longitudinal trend feature. The lateral deviation feature and the vertical trend feature are concatenated and fused to generate a multidimensional cognitive bias vector; The multidimensional cognitive bias vector is mapped and classified using a hybrid attribution model to obtain the corresponding scientific research defect labels.

8. The method according to claim 7, characterized in that, The hybrid attribution model includes a rule engine and a classifier. The hybrid attribution model is used to map and classify the multidimensional cognitive bias vector to obtain corresponding research deficiency labels, including: The rule engine is used to perform violation and anomaly filtering on the multidimensional cognitive bias vector to obtain a compliant target multidimensional cognitive bias vector. For a target multidimensional cognitive bias vector that meets the preset defect judgment conditions, the rule engine directly generates a corresponding scientific research defect label. For a target multidimensional cognitive bias vector that does not meet the preset defect judgment conditions, the target multidimensional cognitive bias vector is input into the classifier. The classifier is used to calculate the matching probability of each scientific research defect label. The label corresponding to the highest matching probability is selected as the final scientific research defect label of the target evaluation sample.

9. The method according to claim 2, characterized in that, The step of matching corresponding intervention strategies based on the research defect labels and generating isomorphic variant verification experiments includes: Retrieve intervention strategies that match the research deficiency tags from a pre-built knowledge graph of intervention strategies, and generate an intervention report based on the matched intervention strategies; Based on the parameters of the original experimental scenario, and under the constraints of preset generation rules and experimental templates, the experimental generation model is invoked to generate isomorphic variant verification experiments and accompanying exercises corresponding to the original experiment.

10. An electronic device, characterized in that, include: The device includes a processor, a memory, and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is in operation, the processor communicates with the memory via the bus, and the processor executes the machine-readable instructions to perform the steps of the method as described in any one of claims 1 to 9.

Images