Intelligent subjective question marking and learning condition analysis system based on writing characteristics

Abstract

The invention provides a subjective question intelligent reviewing and learning condition analysis system based on writing features, and relates to the technical field of education, the system architecture comprises a writing feature acquisition and analysis subsystem, an intelligent reviewing subsystem, and a learning condition analysis and prediction subsystem, and comprehensive innovation and improvement are realized by means of quantum technology. A model constructed by a quantum sensor and calculation enables microcosmic dynamic data to be accurate and observable, time sequence feature depth to be analyzed, a quantum algorithm on the macroscopic aspect has a remarkable effect on font aesthetics and multi-factor association model construction, quantum natural language processing and neural network cooperation, semantic style fusion innovation, knowledge graph reasoning by means of a quantum graph algorithm and the like, and the method has the advantages of being high in practicability and easy to popularize. Various question types can be accurately judged and scientifically guided, quantum hash and federated learning assistance multi-source data fusion is achieved, quantum random walk and an autoregressive moving average model achieve psychological association prediction, learning conditions are accurately informed, physical and psychological and academic co-proceeding of students is effectively intervened and assisted, and the efficiency and value of the system in the education field are greatly enhanced.

Description

Technical Field

[0001] The present invention relates to the field of educational technology, and in particular to a system for intelligent marking of subjective questions and analysis of learning conditions based on writing features. Background Art

[0002] According to a method and system for marking handwritten compositions disclosed in China with the publication number "CN110390032B". The method for marking handwritten compositions includes: obtaining a picture of a handwritten composition to be marked, identifying the picture of the handwritten composition to be marked based on deep learning handwritten text recognition technology to generate the composition text to be marked; performing cluster analysis on each paragraph of the composition text to be marked to determine the problem paragraph; selecting the first sentence of the problem paragraph as the benchmark sentence, and searching for similar sentences similar to the benchmark sentence in the model sentence database, including: calculating the second edit distance between the sentences in the first priority model sentence library, the second priority model sentence library and the third priority model sentence library and the benchmark sentence; selecting M sentences whose second edit distance is less than the second threshold as similar sentences; generating marking results, including recommending similar sentences to students corresponding to the composition text to be marked. The present invention can realize large-scale quantitative processing of handwritten compositions.

[0003] According to a test question marking method and test question marking system disclosed in China with the publication number "CN109064814A". The test question marking method includes: receiving the test question information uploaded by the teacher end, generating an identification code according to the test question information; sending the identification code to the designated terminal, so that the designated terminal obtains and displays the corresponding test question according to the identification code; obtaining the test question answer uploaded by the designated terminal, comparing the test question answer with the pre-stored standard answer, and generating a comparison result; generating marking statistical data of preset statistical dimensions according to the comparison result, and sending the marking statistical data to the teacher end. The present invention pushes the test questions to students through the identification code, is simple to operate, realizes rapid collection of students' answers and automatic marking, obtains the analysis results required by teachers, and greatly reduces the burden on teachers.

[0004] According to a test question review system disclosed in China Publication No. "CN106355176A", it is characterized by comprising a transfer device module for separating and transferring test papers, a photographing module for taking photos of test papers, an information sending module and an information receiving module for transmitting photos to a computer, the transfer device module separates the test papers one by one, and then the photographing module takes photos and transmits the photos obtained by the photos to the computer through the information sending module and the information receiving module, the computer uses the OCR technology recognition module to recognize the pictures and then compares them with the test questions in the large database module, analyzes the right and wrong of the test questions and scores them through the scoring module. The test question review system of the present invention does not need to use a specific answer sheet, has no requirements on the paper of the test paper, has a wide range of applications, saves expensive components such as a cursor answer sheet scanner, and can review multiple-choice questions and simple fill-in-the-blank questions in daily exams or homework, greatly reducing the review burden of teachers.

[0005] The above patent documents and prior art have the following technical problems when used:

[0006] Problem 1: Traditional subjective question marking and learning situation analysis systems are unable to comprehensively, accurately and deeply collect and analyze writing characteristics, and are unable to fully explore the various aspects of student information contained in micro and macro writing characteristics, such as writing habits, thinking patterns, psychological states, and the relationship with cultural background and subject preferences, resulting in insufficient in-depth and comprehensive understanding of students, making it difficult to provide accurate and personalized education services;

[0007] Problem 2: Existing intelligent marking technology is insufficient in comprehensively considering the semantic correctness of answers, writing style characteristics, and the logical fit of subject knowledge systems. It is difficult to achieve accurate judgment and scientific guidance, and cannot efficiently and accurately handle marking tasks that integrate semantic understanding and writing style. In addition, when faced with questions that require rigorous reasoning and open-ended questions, there is a lack of effective evaluation and feedback mechanisms, which limits the accurate evaluation and promotion of students' knowledge mastery and thinking ability.

[0008] Problem three: Conventional academic analysis systems are inefficient and have limited accuracy when integrating multi-source data. It is difficult to deeply explore the internal connections and patterns between the data, and it is impossible to accurately understand the students' academic situation. It is difficult to achieve timely and effective early warning and accurate prediction. There are technical bottlenecks in conducting joint learning of multi-source data to promote the sharing of educational resources while protecting data privacy. It cannot meet the needs of personalized teaching and the coordinated development of students' physical and mental health and academic performance. Summary of the invention

[0009] Technical issues solved

[0010] In view of the shortcomings of the prior art, the present invention provides an intelligent review and learning situation analysis system for subjective questions based on writing features, which solves the following problems:

[0011] 1. The traditional teaching review system is difficult to deeply and comprehensively analyze the writing characteristics, making it difficult to provide accurate and personalized education services;

[0012] 2. The intelligent marking technology is not good at accurately judging multiple factors and responding to different types of questions;

[0013] 3. The conventional learning situation analysis system has problems with incomplete analysis data and difficulty in making accurate predictions and early warnings for anomalies.

[0014] Technical Solution

[0015] To achieve the above objectives, the present invention is implemented through the following technical solutions: a system for intelligent review and learning situation analysis of subjective questions based on writing features, the system architecture includes a writing feature collection and analysis subsystem, an intelligent review subsystem, and a learning situation analysis and prediction subsystem, wherein:

[0016] The writing feature acquisition and analysis subsystem is used to accurately collect multi-dimensional feature information in the student's writing process, and deeply analyze the information such as the student's writing habits, thinking patterns and psychological states behind these features. The writing feature acquisition and analysis subsystem includes a micro-feature acquisition and analysis module and a macro-feature quantification and association module. The micro-feature acquisition and analysis module uses a high-sensitivity quantum sensor to collect micro-dynamic data such as pen tip pressure changes, tilt angles and accelerations, and quickly builds a hand muscle movement model through quantum computing to analyze the student's force stability, stroke coherence and emotional state when writing, and accurately locate the psychological fluctuations or knowledge uncertainty points reflected by abnormal changes in pressure and angles. The macro-feature quantification and association module introduces a multi-dimensional quantitative indicator system, uses quantum machine learning algorithms (such as quantum support vector machines, quantum clustering algorithms) to quantify font aesthetic features (symmetry, balance, etc.) and analyze the association with psychological traits, and uses quantum Bayesian network algorithms to build a writing style and cultural background, subject preference association model to explore the influence of different cultures and subjects on the shaping of writing style and hidden associations;

[0017] The intelligent review subsystem is used to review students' answers to subjective questions comprehensively, accurately and intelligently, comprehensively consider the semantic correctness of the answers, the characteristics of the writing style, and the logical fit with the subject knowledge system, and give objective, fair and constructive review results and feedback. The intelligent review subsystem includes a semantic style fusion review module and an intelligent graph reasoning review module. The semantic style fusion review module uses a quantum natural language processing algorithm (such as quantum latent semantic analysis) to understand the semantics of the answer, combines the quantum neural network to analyze the writing style characteristics, and comprehensively scores based on the connection between the two. According to the font emphasis and semantic accuracy when expressing knowledge points, reasonable weights are given, and the degree of mastery of knowledge points when the font is hesitant is judged and the score is adjusted. The knowledge graph reasoning and review module uses quantum graph algorithm to construct subject knowledge graph, locates the associated logic in the graph according to the knowledge points of the student's answer, and checks the coverage and derivation process of knowledge nodes, such as combing various knowledge chains in physics questions and annotating loopholes, using quantum logic gate circuits to simulate the reasoning process, disassembling the reasoning steps such as mathematical proof questions, checking their standardization and logic, and using quantum search algorithms to evaluate the answer ideas of open questions, and provide score ranges and evaluation suggestions;

[0018] The learning situation analysis and prediction subsystem: through deep integration of multi-source data, it explores the intrinsic connection and rules between writing characteristics and learning behavior, psychological state and academic performance, realizes all-round insight into students' learning situation, timely warning and accurate prediction, and provides a strong basis for personalized teaching and learning support. The learning situation analysis and prediction subsystem includes a multi-source data fusion insight module and a psychological characteristic correlation prediction module. The multi-source data fusion insight module builds a data integration platform, integrates writing characteristics and learning behavior data, uses quantum encryption to ensure data security, and explores the correlation pattern between the two through quantum data mining and machine learning algorithms, such as analyzing classroom notes. It is associated with homework and test performance to discover learning difficulties or advantages corresponding to different writing and behavioral characteristics. With the help of quantum computing to accelerate complex correlation analysis, the psychological characteristic correlation prediction module constructs a writing psychology analysis model and uses the quantum random walk algorithm to analyze the correlation between writing pressure, speed and other characteristics and psychological factors in a high-dimensional feature space, inferring changes in psychological pressure, interest and other states. Combined with historical performance data, it uses quantum time series analysis and machine learning prediction algorithms to build a performance prediction model, predict future performance trends based on writing psychological states and performance trends, and provide warnings and teaching suggestions for different situations to promote the coordinated development of students' physical and mental health and academic performance.

[0019] Preferably, the system also includes a data visualization subsystem for presenting the data and results generated by the writing feature collection and analysis subsystem, the intelligent marking subsystem, and the learning situation analysis and prediction subsystem to teachers, students, and education administrators in an intuitive and visual manner. The subsystem uses quantum rendering technology to improve the clarity and interactivity of the visualization effect, so that users can more easily understand and use the information output by the system, and display the distribution of students' writing features in different dimensions and their correlation with academic performance through quantum three-dimensional visualization.

[0020] Preferably, the system has a self-optimization function. By regularly collecting user feedback data and new learning sample data, the system uses quantum reinforcement learning algorithms to adjust parameters and optimize structures of various models in the system, including writing feature analysis models, intelligent marking models, learning situation prediction models, etc., so as to adapt to the ever-changing educational and teaching environment and student group characteristics, and improve the performance and accuracy of the system. For example, based on students' doubts about the marking results and new syllabus requirements, the scoring rules and semantic understanding strategies in the intelligent marking model are automatically optimized.

[0021] Preferably, when the macro-feature quantification and association module uses a quantum support vector machine to quantify font aesthetic features, the kernel function adopts a quantum Gaussian kernel function, and its parameters are adjusted through a quantum optimization algorithm to improve the accuracy of font aesthetic feature classification. When the association model is constructed using a quantum Bayesian network algorithm, the determination of the prior probability distribution is based on quantum statistical analysis of large-scale sample data, so that the model can more accurately reflect the relationship between writing style and multiple factors.

[0022] Preferably, the quantum neural network in the semantic-style fusion review module adopts a quantum convolutional neural network architecture, and the convolution kernel size and step size of its quantum convolution layer are optimized according to the spatial distribution characteristics of the writing style features. When combining quantum latent semantic analysis to understand the semantics of the answer, the dimension of the semantic vector is reduced through a quantum principal component analysis algorithm to reduce the computational complexity and retain key semantic information, thereby improving the efficiency and accuracy of the comprehensive scoring.

[0023] Preferably, in the process of constructing the subject knowledge graph, the knowledge graph reasoning and review module uses a quantum graph algorithm to traverse the knowledge nodes and discover relationships, so as to enhance the integrity and accuracy of the knowledge graph construction. When using quantum logic gate circuits to simulate the reasoning process, quantum error correction code technology is used to reduce the error rate in the quantum computing process, thereby ensuring the reliability of the reasoning step checks.

[0024] Preferably, when fusing writing features and learning behavior data, the multi-source data fusion insight module uses a quantum hashing algorithm to pre-process the data to improve the speed and accuracy of data fusion, and when using quantum data mining and machine learning algorithms to explore association patterns, a quantum federated learning algorithm is used, so that when multi-source data are distributed in different storage locations, joint learning can still be effectively performed, protecting data privacy while mining valuable association information. When the psychological feature association prediction module constructs a writing psychological analysis model, the step size and number of iterations of the quantum random walk algorithm are adaptively adjusted according to the scale and complexity of the writing feature data to ensure efficient and accurate analysis of the association between features and psychological factors in a high-dimensional feature space, and when constructing a performance prediction model, the quantum time series analysis algorithm uses a quantum autoregressive moving average model (QARIMA) to better capture the time series characteristics and trend changes of the performance data.

[0025] Beneficial Effects

[0026] The present invention provides a system for intelligent reviewing of subjective questions and analyzing learning conditions based on writing features. It has the following beneficial effects:

[0027] 1. The system adopted by the present invention has the functions of deep analysis and comprehensive quantification in terms of writing characteristics. In terms of writing characteristics, quantum technology helps to capture microscopic characteristics with ultra-precision and deep analysis. The hand muscle movement model constructed by high-sensitivity quantum sensors combined with quantum computing can perceive dynamic data such as pen tip pressure, tilt angle and acceleration in detail, and accurately interpret the students' force stability, stroke continuity, and subtle changes in emotions and knowledge mastery when writing. At the same time, it efficiently analyzes the stroke timing characteristics, explores the coherence of thinking and the fluency of knowledge integration. In terms of macroscopic characteristics, quantum machine learning algorithms such as quantum support vector machines and quantum clustering algorithms, as well as quantum Bayesian network algorithms, respectively realize the precise quantification of font aesthetic characteristics and the in-depth construction of writing style and multi-factor association model, comprehensively revealing the rich student information behind the writing style, and laying a solid foundation for subsequent intelligent review and learning situation analysis.

[0028] 2. The system adopted by the present invention has the functions of accurate judgment and scientific guidance in intelligent marking. In terms of intelligent marking, quantum technology helps to innovate the marking mode that integrates semantics and style. The quantum natural language processing algorithm and the quantum neural network work together to deeply analyze the writing style characteristics while understanding the semantics of the answer, and intelligently allocate weights according to the font presentation and semantic accuracy of the knowledge point expression, optimize the comprehensive scoring mechanism, and improve the marking efficiency and accuracy. The knowledge graph reasoning marking uses the quantum graph algorithm to build a complete subject knowledge graph, accurately locate the association logic of students' answers in the knowledge system, use quantum logic gate circuits to simulate the reasoning process and combine quantum error correction code technology to ensure the high reliability of reasoning checks. For open questions, it relies on quantum search algorithms to quickly explore multiple answer ideas, provide reasonable scoring ranges and constructive evaluation suggestions, effectively improve the scientificity and comprehensiveness of marking, and promote students' knowledge mastery and thinking expansion.

[0029] 3. The system used in the present invention can provide accurate insights and effective interventions in academic situation analysis. In terms of academic situation analysis, quantum technology helps to integrate multi-source data and predict the association between psychological characteristics. The quantum hash algorithm pre-processes writing characteristics and learning behavior data, accelerates the fusion process and accurately mines the association patterns between the two, and quickly locates students' learning difficulties and advantages. The quantum federated learning algorithm realizes multi-source data joint learning while protecting data privacy, promotes educational resource sharing and collaborative optimization, and the psychological characteristic association prediction module adaptively analyzes the association between writing characteristics and psychological factors with the help of the quantum random walk algorithm, accurately infers changes in students' psychological states, and combines the performance prediction model constructed with the quantum autoregressive moving average model to accurately predict the trend of students' performance, thereby providing a strong basis for personalized teaching and timely psychological intervention, and promoting the coordinated progress of students' physical and mental health and academic progress. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 is a system architecture diagram of the present invention;

[0031] Figure 2 It is a diagram of the system operation steps of the present invention;

[0032] Figure 3 It is a line graph showing the changes of pressure, angle and acceleration over time in the microscopic feature acquisition and analysis module of the present invention;

[0033] Figure 4 A scatter plot of the font aesthetic feature classification results in the macro feature quantification and association module of the present invention;

[0034] Figure 5 A three-dimensional display diagram of macro-feature clustering results written in the macro-feature quantification and association module of the present invention;

[0035] Figure 6A distribution diagram of documents in a low-dimensional semantic space in the semantic style fusion review module of the present invention;

[0036] Figure 7 A comparison diagram of the original writing style feature image and the image after convolution in the semantic style fusion review module of the present invention;

[0037] Figure 8 It is a probability amplitude distribution diagram of knowledge nodes after quantum random walk in the knowledge graph reasoning and review module of the present invention;

[0038] Fig. 9 It is a quantum state probability amplitude distribution diagram after the NOT gate operation in the knowledge graph reasoning and review module of the present invention;

[0039] Fig.10 It is a three-dimensional graph of the clustering results of the writing characteristics and learning behavior data of the present invention. DETAILED DESCRIPTION

[0040] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. Specific embodiment one:

[0042] like Figure 1-10 As shown in the figure, the system of intelligent marking and learning situation analysis based on writing features for subjective questions includes a writing feature collection and analysis subsystem, an intelligent marking subsystem, and a learning situation analysis and prediction subsystem, among which:

[0043] Writing feature collection and analysis subsystem: used to accurately collect multi-dimensional feature information of students' writing process, and deeply analyze the information behind these features, such as students' writing habits, thinking patterns, and psychological states. The writing feature collection and analysis subsystem includes a micro-feature collection and analysis module and a macro-feature quantification and association module;

[0044] The micro-feature acquisition and analysis module uses highly sensitive quantum sensors to collect micro-dynamic data such as pen tip pressure changes, tilt angles, and accelerations. Through quantum computing, it quickly builds a hand muscle movement model to analyze students' force stability, stroke continuity, and emotional state when writing, and accurately locates psychological fluctuations or knowledge uncertainty points reflected by abnormal changes in pressure and angles. The hand muscle movement model built with highly sensitive quantum sensors and quantum computing can capture micro-dynamic data such as pen tip pressure, tilt angles, and accelerations with extremely high precision. This makes the analysis of students' force stability and stroke continuity when writing reach an unprecedented level of detail. For example, it can accurately perceive pressure changes at the micronewton level, thereby accurately locating students' writing process. Subtle emotional fluctuations or moments of uncertainty about specific knowledge points. This kind of deep insight at the micro level helps teachers to fully understand students' psychological state when writing and their instantaneous reaction to knowledge mastery, providing a valuable reference for personalized teaching. Micro feature analysis based on quantum computing can process massive amounts of time series data, quickly and accurately analyze time series features such as the connection sequence between strokes, pause duration and frequency, and writing speed change curve. Compared with traditional analysis methods, not only is the speed greatly improved, but also deeper information on thinking coherence and knowledge integration fluency can be excavated. For example, when analyzing students writing a complex argumentative article, it can clearly identify the students' thinking jams when switching between different paragraphs or knowledge points, providing accurate basis for targeted tutoring.

[0045] The macro-feature quantification and association module introduces a multi-dimensional quantitative indicator system, uses quantum machine learning algorithms (such as quantum support vector machines and quantum clustering algorithms) to quantify font aesthetic features (symmetry, balance, etc.) and analyze the association with psychological traits, and uses quantum Bayesian network algorithms to build a writing style and cultural background, subject preference association model, and explore the influence of different cultures and subjects on the shaping of writing style and hidden associations. When the macro-feature quantification and association module uses quantum support vector machines to quantify font aesthetic features, the kernel function uses quantum Gaussian kernel functions, and its parameters are adjusted through quantum optimization algorithms to improve the accuracy of font aesthetic feature classification. When using quantum Bayesian network algorithms to build association models, the determination of prior probability distribution is based on quantum statistical analysis of large-scale sample data, so that the model can more accurately reflect the relationship between writing style and multiple factors. Quantum machine learning algorithms (such as quantum support vector machines and quantum clustering algorithms) are used to The aesthetic features of fonts (symmetry, balance, etc.) are quantified, and the accuracy of font aesthetic feature classification is significantly improved with the help of quantum Gaussian kernel functions and their optimized parameters. This helps to more accurately characterize students' writing style characteristics. For example, writing samples with similar traditional features but slightly different aesthetic styles can be more accurately distinguished, providing a more reliable data basis for studying the relationship between writing style and students' personality traits. The quantum Bayesian network algorithm is used to construct an association model between writing style and cultural background and subject preference. The prior probability distribution is determined based on quantum statistical analysis of large-scale sample data, which can deeply explore the influence of different cultures and disciplines on the shaping of writing style and hidden associations. For example, the potential connection between the regularity of students' handwriting strokes in the context of oriental culture and the inheritance of Chinese character culture can be discovered, as well as the mechanism by which science learning promotes the conciseness and symbolic use of students' writing, providing strong support for cross-cultural and cross-disciplinary educational research and teaching strategy formulation.

[0046] Intelligent marking subsystem: used to conduct comprehensive, accurate and intelligent marking of students' answers to subjective questions, comprehensively considering multiple factors such as the semantic correctness of the answers, the characteristics of the writing style, and the logical fit with the subject knowledge system, and giving objective, fair and constructive marking results and feedback. The intelligent marking subsystem includes a semantic style fusion marking module and an intelligent graph reasoning marking module.

[0047] The semantic-style fusion review module uses quantum natural language processing algorithms (such as quantum latent semantic analysis) to understand the semantics of the answer, combines quantum neural networks to analyze writing style characteristics, and comprehensively scores based on the connection between the two. It gives reasonable weights based on the font emphasis and semantic accuracy when the knowledge points are expressed, and judges the degree of mastery of the knowledge points when the font is hesitant and adjusts the score. The quantum neural network in the semantic-style fusion review module uses a quantum convolutional neural network architecture. The convolution kernel size and step size of its quantum convolution layer are optimized according to the spatial distribution characteristics of the writing style characteristics. When combining quantum latent semantic analysis to understand the semantics of the answer, the dimension of the semantic vector is reduced through the quantum principal component analysis algorithm to reduce the computational complexity and retain key semantic information, thereby improving the efficiency and accuracy of the comprehensive score. The combination of quantum natural language processing algorithms (such as quantum latent semantic analysis) and quantum neural networks realizes the organic integration of semantic understanding and writing style feature analysis. In the review process, it can be based on the font emphasis and semantic accuracy when the knowledge points are expressed. Accuracy is given a reasonable weight. For example, for core knowledge points that are highlighted in larger fonts and bold, if the semantic expression is accurate and clear, a higher scoring weight will be given, otherwise it will be adjusted accordingly. This integrated review method not only takes into account the correctness of the answer content, but also takes into account the students' emphasis on knowledge and their confidence in expression, making the review results more objective, fair and insightful. When analyzing writing style features, the quantum convolutional neural network architecture can more accurately capture the spatial distribution characteristics of writing style features by optimizing the convolution kernel size and step size of the quantum convolution layer, such as the correlation between the regional distribution of font size changes and the importance of knowledge points. At the same time, the quantum principal component analysis algorithm performs dimensionality reduction processing on the semantic vector, which reduces the computational complexity while retaining key semantic information, greatly improving the efficiency and accuracy of the comprehensive scoring. For example, when reviewing a subjective question containing multiple paragraphs of discussion, it can quickly and accurately evaluate the quality of each paragraph, and comprehensively derive an overall score, while providing students with detailed feedback and suggestions on semantics and writing style.

[0048] The knowledge graph reasoning and review module uses quantum graph algorithms to construct subject knowledge graphs, locates associated logic in the graph based on students' answers, and checks the coverage and derivation process of knowledge nodes. For example, in physics questions, it sorts out various knowledge chains and annotates loopholes, uses quantum logic gate circuits to simulate the reasoning process, disassembles mathematical proof questions and other reasoning steps, checks their standardization and logic, and uses quantum search algorithms to evaluate open-ended question answer ideas, provides score ranges and evaluation suggestions. In the process of constructing subject knowledge graphs, the knowledge graph reasoning and review module uses quantum random walk algorithms to traverse knowledge nodes and discover relationships to enhance the integrity and accuracy of knowledge graph construction, and uses quantum logic gate circuits to simulate the reasoning process. When programming, quantum error correction code technology is used to reduce the error rate in the quantum computing process and ensure the reliability of the reasoning step check. The quantum graph algorithm uses a quantum random walk algorithm to construct a subject knowledge graph, which can more comprehensively traverse and discover knowledge nodes and their relationships, significantly enhancing the completeness and accuracy of the knowledge graph construction. When marking, the associated logic can be more accurately located in the graph based on the knowledge points of the students' answers. For example, in the knowledge graph of physics, for complex problems involving multiple physical concepts and principles, the complete knowledge chain can be quickly sorted out, and the student answers can be accurately checked to see if they cover all the key nodes and their derivation processes. The loopholes in the student's knowledge system can be discovered in a timely manner and detailed annotations can be given, which improves the depth and quality of the marking.

[0049] Quantum logic gate circuits are used to simulate the reasoning process, and quantum error correction code technology is used to reduce the error rate in the quantum computing process, ensuring the reliability of the reasoning step checks. In the marking of mathematical proof questions that require rigorous logical reasoning, the standardization and logic of each step of reasoning can be accurately judged, and marking bias caused by calculation errors or logical misjudgments can be effectively avoided. At the same time, the quantum search algorithm evaluates the answer ideas for open-ended questions, and can quickly explore multiple possible answer directions, providing students with reasonable score ranges and inspiring evaluation suggestions, encouraging students' innovative thinking and diversified expressions, and improving the flexibility and scientific nature of marking.

[0050] Learning situation analysis and prediction subsystem: Through deep fusion of multi-source data, the intrinsic connection and rules between writing characteristics and learning behavior, psychological state and academic performance are explored to achieve all-round insight into students' learning situation, timely warning and accurate prediction, and provide a strong basis for personalized teaching and learning support. The learning situation analysis and prediction subsystem includes a multi-source data fusion insight module and a psychological characteristic correlation prediction module;

[0051] The multi-source data fusion insight module builds a data integration platform, integrates writing features and learning behavior data, uses quantum encryption to ensure data security, and explores the correlation pattern between the two through quantum data mining and machine learning algorithms, such as analyzing the correlation between classroom notes and homework and test performance, discovering the learning difficulties or advantages corresponding to different writing and behavior characteristics, and accelerating complex correlation analysis with the help of quantum computing. When integrating writing features and learning behavior data, the multi-source data fusion insight module uses the quantum hashing algorithm to pre-process the data to improve the speed and accuracy of data fusion. When using quantum data mining and machine learning algorithms to explore correlation patterns, the quantum federated learning algorithm is used, so that when multi-source data are distributed in different storage locations, joint learning can still be effectively carried out, protecting data privacy while mining valuable correlation information. The quantum hashing algorithm pre-processes the writing features and learning behavior data, greatly improving the speed and accuracy of data fusion. When the performance of students' homework and examinations is correlated, it is possible to more quickly discover the learning difficulties or advantages corresponding to different writing and behavioral characteristics. For example, the quantum hash algorithm can be used to quickly screen out the student group with frequent handwriting corrections and long homework completion time, and further analyze their learning behavior data to accurately locate the possible knowledge understanding or application difficulties, provide teachers with timely intervention basis, and improve the pertinence and timeliness of teaching. The quantum federated learning algorithm can still effectively perform joint learning when multi-source data is distributed in different storage locations, and valuable related information can be mined while protecting data privacy. This allows different schools and educational institutions to share learning situation analysis experience and results without leaking students' privacy data, promote the optimal allocation of educational resources and the realization of educational equity. For example, schools in different regions can jointly participate in learning situation analysis projects based on quantum federated learning, update information by sharing model parameters, and jointly improve the understanding of the learning situation of student groups and the formulation of teaching strategies.

[0052] The psychological feature association prediction module constructs a writing psychology analysis model and uses the quantum random walk algorithm to analyze the association between writing pressure, speed and other features and psychological factors in a high-dimensional feature space, infers changes in psychological pressure, interest and other states, and combines historical performance data to build a performance prediction model using quantum time series analysis and machine learning prediction algorithms. It predicts future performance trends based on the writing psychological state and performance trends, provides early warnings and provides teaching suggestions for different situations, and promotes the coordinated development of students' physical and mental health and academic performance. When constructing the writing psychology analysis model, the step size and number of iterations of the quantum random walk algorithm are adaptively adjusted according to the scale and complexity of the writing feature data to ensure efficient and accurate analysis of the association between features and psychological factors in a high-dimensional feature space. When constructing the performance prediction model, the quantum time series analysis algorithm uses the quantum autoregressive moving average model (QARIMA) to better capture the time series characteristics and trend changes of the performance data. In the psychological feature association prediction module, the step size and number of iterations of the quantum random walk algorithm are adaptively adjusted according to the scale and complexity of the writing feature data. The complexity is adaptively adjusted to ensure efficient and accurate analysis of the association between writing pressure, speed and other characteristics and psychological factors in the high-dimensional feature space, and can accurately infer students' psychological stress levels, fluctuations in learning interests, changes in self-efficacy, and emotional state fluctuations. For example, when facing exam pressure, by analyzing the sudden changes in students' writing speed and the fluctuations in stress data, the changes in students' psychological state can be accurately judged, providing timely and accurate information support for schools to carry out psychological counseling and intervention. The quantum autoregressive moving average model (QARIMA) is used when constructing the score prediction model, which can better capture the time series characteristics and trend changes of score data, and combine the writing psychological state with the historical score trend to more accurately predict the future performance of students. For example, for a student whose writing psychological state is gradually stable and whose historical scores are on an upward trend, it can accurately predict the extent of his or her score improvement in future exams, and provide teachers with personalized teaching resource recommendations and learning task arrangement suggestions to promote the coordinated development of students' physical and mental health and academic performance, and realize precise education and teaching management and personalized learning guidance.

[0053] The system also includes a data visualization subsystem, which is used to present the data and results generated by the writing feature collection and analysis subsystem, the intelligent marking subsystem, and the learning situation analysis and prediction subsystem to teachers, students, and education administrators in an intuitive and visual manner. The subsystem uses quantum rendering technology to improve the clarity and interactivity of the visualization effect, so that users can more easily understand and use the information output by the system. The quantum three-dimensional visualization shows the distribution of students' writing features in different dimensions and their correlation with academic performance.

[0054] The system has a self-optimization function. By regularly collecting user feedback data and new learning sample data, it uses quantum reinforcement learning algorithms to adjust parameters and optimize the structure of various models in the system, including writing feature analysis models, intelligent marking models, and learning situation prediction models, so as to adapt to the ever-changing educational and teaching environment and student group characteristics, and improve the performance and accuracy of the system. For example, based on students' doubts about the marking results and new teaching syllabus requirements, the scoring rules and semantic understanding strategies in the intelligent marking model are automatically optimized.

[0055] In this system architecture, quantum technology has been used to achieve all-round innovation and improvement. At the writing feature level, the model constructed by quantum sensors and computing makes microscopic dynamic data accurately observable and time series features deeply analyzable. At the macro level, quantum algorithms have achieved remarkable results in the construction of font aesthetics and multi-factor correlation models, laying the foundation for the overall system. In the intelligent marking link, quantum natural language processing and neural networks collaborate, semantic style fusion innovation, knowledge graph reasoning with the help of quantum graph algorithms, etc., the reasoning is rigorous, and all kinds of questions can be accurately judged and scientifically guided. In terms of learning situation analysis, quantum hashing and federated learning help to integrate multi-source data, quantum random walks and autoregressive moving average models realize psychological correlation prediction, accurately understand the learning situation, and effectively intervene to help students' physical and mental health and academic progress in a coordinated manner, greatly enhancing the effectiveness and value of the system in the field of education. Specific embodiment 2:

[0057] like Figure 1-10 As shown, according to the contents in the above specific embodiments, the following contents are further disclosed:

[0058] The specific contents of each algorithm in the above system architecture are as follows:

[0059] Using highly sensitive quantum sensors to collect data and build hand muscle movement models

[0060] Implementation steps:

[0061] First, a highly sensitive quantum sensor is used to collect microscopic dynamic data such as pen tip pressure changes, tilt angles, and acceleration in real time. The acquisition of these data is the basis for subsequent analysis. For example, during the writing process, the sensor will record the current pressure value, tilt angle value, and acceleration value of the pen tip at a certain time interval (such as every millisecond).

[0062] Next, we use quantum computing methods to build a hand muscle movement model based on the collected dynamic data. Quantum computing, with its unique quantum bit (qubit) characteristics, can process more complex state information. Compared with traditional computing, it can operate on multiple states at the same time, thereby speeding up the model building speed. For example, quantum bits are used to represent the data state combination corresponding to different hand muscle movements, and then these state information are integrated to build a model that reflects the hand muscle movement.

[0063] Operation logic: The movement of hand muscles during writing directly affects the state of the pen tip, and these microscopic dynamic data are the external manifestations of hand muscle movement. Quantum computing uses the superposition and entanglement characteristics of quantum states to conduct a comprehensive analysis of a large amount of data collected at different times, and explore the internal connections among them. The model constructed in this way can accurately reflect how the hand muscles move to complete writing movements, such as which muscles will cause changes in pen tip pressure, what kind of inclination angle changes correspond to hand movements in different directions, etc. It can accurately locate the psychological fluctuations or knowledge uncertainty points reflected by abnormal changes in pressure and angle. For example, when students are writing content related to a certain knowledge point, if the pen tip pressure suddenly increases and the stroke continuity is abnormal, the model combined with subsequent analysis can infer that it may be due to a lack of solid grasp of this knowledge point, some inner tension and other psychological state changes, which will help teachers to have a deeper understanding of students' mastery of knowledge.

[0064] Quantum support vector machine, used for macro feature quantification and font aesthetic feature quantification in the association module,

[0065] For the standard support vector machine for binary classification, the basic mathematical formula is as follows:

[0066]

[0067] Among them, w is the weight vector (representing the direction of the normal vector dividing the hyperplane), b is the bias term, and x i is the input sample feature vector (here it is the feature vector related to font aesthetics, such as the vector composed of the symmetry, balance and other feature values ​​of the font), y i is the corresponding sample category label (for example, you can set the font with high aesthetics to +1, the font with low aesthetics to -1, etc.), n is the number of samples, represents the inner product operation of the vector, ‖w‖ 2 Represents the square of the bi-norm of the weight vector (that is, the sum of the squares of the elements of the vector, the purpose of which is to control the interval of the hyperplane to maximize the classification effect);

[0068] Implementation steps:

[0069] The first step is to collect a large amount of font sample data with annotations (font aesthetic category labels) and extract the font aesthetic feature vector and its category label y corresponding to each sample. i ;

[0070] The second step is to substitute these data into the above objective function and constraints, and use an optimization algorithm (such as the commonly used sequence minimum optimization algorithm, which may be accelerated in quantum scenarios by combining quantum optimization algorithms) to solve the values ​​of w and b, and continuously adjust the weight vector and bias term so that the divided hyperplane can better separate fonts of different aesthetic levels.

[0071] The third step is to actually quantify and classify the aesthetic features of the new font by substituting its corresponding feature vector into the classification function f(x)=w·x+b (the symbol function of category judgment and other details are omitted here for simplified description) determined by the trained w and b to determine which category of aesthetically pleasing font it belongs to;

[0072] Improving the accuracy of classifying font aesthetic features can more accurately distinguish fonts of different levels of beauty and styles, thereby more deeply exploring the relationship between font aesthetic features and students' psychological characteristics. For example, it is found that students with high levels of beautiful handwriting may pay more attention to details during the learning process and other psychological characteristics, providing a strong basis for subsequent learning situation analysis.

[0073] Quantum clustering algorithm is used for macro feature quantification and correlation analysis in the association module, combining the idea of ​​general clustering algorithm with quantum characteristics.

[0074] Implementation steps:

[0075] First, determine the objects to be clustered, that is, collect relevant writing feature data, such as taking the overall macroscopic feature data of students' handwriting (which may include multiple dimensions such as font size distribution, character spacing, etc., as a comprehensive feature vector representation) as the data set to be clustered;

[0076] Then, the cluster centers are initialized. Some data points can be randomly selected as the initial cluster centers (in the quantum clustering algorithm, the selection or representation of these initial points may be more efficient, such as using quantum states to store).

[0077] Next, the distance from each data point to each cluster center is calculated (in a quantum scenario, distance calculation may be completed more quickly by utilizing the superposition and interference characteristics of quantum states, such as by reflecting the distance relationship through the evolution of quantum states), and the data points are divided into the category of the closest cluster center according to the distance;

[0078] Then recalculate the cluster center based on the newly divided data points (for example, take the mean of the feature vectors of the data points of the same category to update the cluster center. Quantum computing can accelerate the calculation process of the mean, etc.), and repeat the above steps until the cluster center no longer changes significantly or reaches the set number of iterations or other stopping conditions.

[0079] Operation logic: Based on the similarity between data points (reflected by distance and other metrics), similar writing feature data are clustered into one category. By continuously adjusting the cluster center and re-dividing the data points, the similarity of data points in the same category is as high as possible, and the similarity of data points between different categories is as low as possible. The quantum clustering algorithm utilizes the unique computing and storage characteristics of quantum states. For example, quantum bits can represent multiple states at the same time. In the process of calculating distances and updating cluster centers, multiple situations can be processed in parallel to improve clustering efficiency and effectiveness. The clustering of different writing styles and features can be discovered from a macro perspective. For example, students with similar macroscopic features of fonts can be classified into one category, and then the commonalities that may exist among these students in terms of psychological characteristics, learning habits, etc. can be analyzed to assist in exploring the correlation between writing features and other factors, and provide a basic data division basis for the subsequent construction of association models.

[0080] Quantum Bayesian network algorithm is used for macro-feature quantification and correlation module to build writing style and multi-factor correlation model.

[0081] In the Bayesian network, for node X, which can represent variables such as writing style, given its parent node set Pa(X), its conditional probability distribution formula is as follows:

[0082]

[0083] in,

[0084] P(X|Pa(X)) represents the probability of node X occurring when the value of the parent node set Pa(X) is known, which reflects the conditional dependency between nodes. For example, when the student’s cultural background (parent node) is known, the probability of his / her writing style (node ​​X) showing a certain characteristic.

[0085] P(X, Pa(X)) represents the joint probability, which indicates the probability of node X and its parent node set Pa(X) occurring simultaneously. It covers the probability description of all possible combinations between the two.

[0086] P(Pa(X)) is the probability of the parent node set itself, which is the basis for calculating conditional probability and is determined by statistics or known prior information;

[0087] The joint probability distribution of the entire Bayesian network can be obtained by multiplying the conditional probability distribution of each node (assuming that the network structure is a directed acyclic graph and the dependency relationship between nodes conforms to the Bayesian network structure), that is:

[0088]

[0089] Here X1, X2, ..., X n are all the node variables in the network (such as writing style, cultural background, subject preference, etc.), P(X1, X2, ..., X n ) is the joint probability distribution of the entire Bayesian network, reflecting the probability of all node variables appearing in various value combinations at the same time. It is calculated by multiplying the conditional probability distribution of each node based on its parent node, reflecting the complex probability relationship between variables in the entire network structure.

[0090] The network structure is constructed based on the logical relationship between each factor, and the dependency relationship between these factors is described in the form of probability. The conditional probability distribution is adjusted through prior probability and sample data learning, and then probabilistic reasoning can be performed when partial information is known to obtain the probability of other related factors. The quantum Bayesian network algorithm uses the special properties of the quantum state to accelerate the processes of probability calculation, statistical analysis, and network training optimization, and more accurately reflects the complex correlation between writing style and cultural background, subject preference and other factors. It can construct a correlation model between writing style and cultural background and subject preference, and explore the influence of different cultures and subjects on the shaping of writing style and hidden correlations. For example, it is found that students from certain cultural backgrounds prefer a certain font feature in their writing style, or that students with a certain subject learning preference have specific norms in their writing style. This helps to more comprehensively understand the influencing factors behind students' writing and provide a reference basis for personalized teaching.

[0091] Quantum latent semantic analysis in the semantic style fusion review module is used to understand the semantics of the answer.

[0092] In traditional latent semantic analysis, singular value decomposition is mainly involved. Suppose we have a document-term matrix A m ×n, where m is the number of documents and n is the number of terms. The singular value decomposition formula is:

[0093] A=U∑V T

[0094] Among them, U m×n is the left singular vector matrix, ∑ m×n is a diagonal matrix, and the elements on the diagonal are singular values ​​σ i (i=1,2,…,min(m,n)) and σ1≥σ2≥…≥σ min(m，n) ≥0,V n×nis the right singular vector matrix, T represents the transpose operation of the matrix, V T (or ) is the transpose of the right singular vector matrix, which is used to construct the complete singular value decomposition (or similar decomposition under the concept of quantization) equation relationship to accurately extract semantic information from the matrix;

[0095] In quantum latent semantic analysis, the representation of quantum states and related quantum computing operations are involved. However, there is no completely unified and specific quantization formula yet. We can use the classical SVD principle to process and analyze data under the quantum computing framework, such as using quantum bits to represent the feature states of documents or terms. Here, the quantized matrix is ​​represented as Q A , has a similar decomposition form:

[0096]

[0097] Where Q U , Q ∑ , Q V They are the corresponding matrix representations after quantization, whose elements are quantum state-related representations, with dimensions similar to those in the classical case, but different properties of the internal elements, based on quantum information carriers such as quantum bits.

[0098] Operation logic: By performing singular value decomposition (or quantization decomposition) on the document-term matrix, high-dimensional sparse text data is converted into vector representation in a low-dimensional semantic space, and the potential semantic structure behind the text is mined. In this low-dimensional space, documents or terms with similar semantics will be close in the vector space, so vector operations and measurements can be used for semantic analysis. Quantum latent semantic analysis uses the characteristics of quantum computing, such as the parallel processing capability of quantum bits, to process quantum state representations of multiple terms or documents at the same time, accelerating the decomposition process and the extraction and analysis of semantic information. Compared with classical methods, it is more efficient in large-scale text analysis. This method has potential efficiency advantages in data processing and complex semantic mining. It can understand the semantics of answers and mine deep semantic information in answers, rather than just being limited to surface term matching. For example, it can detect students using different expressions but the same semantics to answer knowledge points, thus avoiding misjudgment of the correctness of answers due to differences in expressions. Through the high efficiency of quantum latent semantic analysis (when processing large-scale data), it can quickly process the answers to subjective questions of many students, improve the speed and accuracy of grading, and better adapt to the needs of semantic analysis of answers to subjective questions of different subjects and types, providing a more reliable semantic basis for comprehensive scoring.

[0099] The quantum neural network in the semantic style fusion review module is used to analyze writing style characteristics:

[0100] The quantum convolutional layer operation in the quantum convolutional neural network can be compared to the convolutional layer formula of the classical convolutional neural network, but it involves the operation of quantum states. Assume that the image represented by the input quantum state (which can be compared to a certain two-dimensional representation of writing style characteristics, such as the visualization of the spatial distribution characteristics of strokes) is |ψ in >, the quantum convolution kernel is |κ>, and the quantum convolution operation can be expressed as:

[0101]

[0102] Among them, s represents the displacement (translation) parameter of the convolution kernel, C(s) is the coefficient related to the displacement, |k s > is the quantum convolution kernel state after displacement;

[0103] In quantum convolutional neural networks, there are also operations such as pooling layers and fully connected layers. The calculation of the fully connected layer is similar to the matrix multiplication form of the classical neural network, but the elements are quantum states and related quantum computing operations. For example, for the quantum fully connected layer, assuming that the input quantum state vector is The weight matrix is ​​|W>, the bias quantum state is |b>, and the output quantum state is It can be expressed as:

[0104]

[0105] Among them, the addition and multiplication here are based on the corresponding operations of quantum states, which are essentially different from classical numerical operations. For example, the superposition and entanglement of quantum states will be involved in these calculation processes.

[0106] Operation logic: The quantum neural network extracts local features of writing style characteristics through the quantum convolution layer, uses the quantum convolution kernel to scan the input data in the quantum state space, the pooling layer aggregates and reduces the features, and the fully connected layer integrates and maps the extracted features, and finally outputs the quantum state representation of the writing style characteristics. In the whole process, based on the parallel processing capability of quantum bits, the superposition and entanglement characteristics of quantum states, the quantum neural network can simultaneously process multiple feature dimensions and complex feature relationships. Compared with classical neural networks, it has stronger expression ability and analysis efficiency when processing complex and somewhat ambiguous features such as writing style. Through the training process, parameters such as the quantum weight matrix are continuously adjusted, so that the network can learn the relationship between different writing style characteristics and review results, so as to accurately analyze the writing style characteristics and provide valuable information for the comprehensive score.

[0107] Beneficial effects: It can analyze writing style features more precisely. For example, it can accurately identify style features such as the stroke coherence and force stability when students are writing. And through the powerful feature learning ability of the quantum neural network, it can discover some information related to learning attitudes, knowledge mastery, etc. hidden in the writing style. For example, a relatively chaotic writing style may imply that the student is in a hurry when answering questions or lacks confidence in knowledge mastery. By combining quantum latent semantic analysis to understand the semantics of answers, it can comprehensively score based on the connection between the two, improve the comprehensiveness and accuracy of marking, better evaluate the quality of students' subjective question answers, and provide more targeted feedback for students and more accurate teaching evaluation basis for teachers.

[0108] The quantum graph algorithm in the knowledge graph reasoning marking module is used to construct a subject knowledge graph:

[0109] Mathematical formula (taking the quantum random walk algorithm as an example): The mathematical description of quantum random walk is relatively complex. Briefly speaking, on a graph G=(V, E) (where V is the vertex set and E is the edge set), the state of quantum random walk can be represented by a quantum state |ψ t > representing at time t:

[0110] |ψ t+1 > = U|ψ t >

[0111] where U is the unitary evolution operator of quantum random walk, which is related to the structure of the graph (the connection relationship between vertices) and the evolution rule of the quantum state. Its specific form can be expressed as (taking a simple discrete-time quantum random walk as an example):

[0112]

[0113] Here A uv is the element of the adjacency matrix of graph G, indicating whether there is an edge connection between vertex u and vertex v (if there is an edge, then A uv = 1; if there is no edge, then A uv = 0), |v><u| is the projection operator of the quantum state, and S is the local unitary transformation operator acting on the quantum state (such as a combination of basic quantum gate operations like Pauli matrices, used to realize the transfer and state change of the quantum state between vertices).

[0114] Operation logic: Based on the structure of the graph and the evolution rules of the quantum state, the quantum random walk starts from the initial state, and at each step, the quantum state is transferred and changed between the vertices of the knowledge graph according to the unitary evolution operator. Through multiple iterations, the quantum state can simultaneously explore multiple paths and node relationships in a quantum parallel processing manner, and use the superposition and entanglement characteristics of the quantum state to mine the deep structural information and hidden associations in the knowledge graph. Compared with the traditional random walk algorithm, the quantum random walk can traverse the knowledge graph more efficiently and discover more complex relationship patterns, because it is not limited by serial processing in classical computing and can consider multiple possibilities at the same time, thereby improving the completeness and accuracy of the knowledge graph construction.

[0115] Beneficial effects: Enhance the integrity and accuracy of knowledge graph construction. In the construction of subject knowledge graphs, various relationships between knowledge points can be discovered more comprehensively. For example, in the subject of physics, the various knowledge chains from basic concepts to complex physical phenomena can be sorted out more accurately, and some knowledge point associations that are easily overlooked under traditional methods can be discovered. This helps to more accurately check the coverage and derivation process of knowledge nodes in students' answers based on knowledge graphs when marking students' answers to subjective questions, discover loopholes and logical errors in students' knowledge mastery, and lay the foundation for providing more accurate annotations and feedback, thereby improving the quality and effectiveness of teaching evaluation and promoting students' in-depth understanding and mastery of the subject knowledge system.

[0116] Quantum logic gate circuit in the knowledge graph reasoning review module (used to simulate the reasoning process)

[0117] Mathematical formula (taking simple quantum logic gate operation as an example): For example, for a single quantum bit NOT gate (quantum NOT gate) operation, its matrix is ​​represented as:

[0118] If the input quantum state is |ψ〉=α|0〉+β|1〉 (where α and β are complex numbers, satisfying |α| 2 +|β| 2 =1), the output quantum state after the NOT gate operation |ψ ′ >For:

[0119]

[0120] For multi-qubit logic gate operations, such as the CNOT gate (controlled NOT gate), if there are two qubits |ψ1|ψ2>, where |ψ1>=α1|0>+β1|1>, |ψ2>=α2|0>+β2|1>, the matrix of the CNOT gate is expressed as:

[0121]

[0122] When |ψ1> is the control bit and |ψ2> is the target bit, the output state after the CNOT gate operation is:

[0123]

[0124] Operational logic: The quantum logic gate circuit maps the reasoning process into a combination of quantum logic gate operations, utilizes the superposition and entanglement characteristics of quantum states and the operation rules of logic gates to process the encoded student answer information. At the quantum bit level, multiple possible reasoning branches and states are processed simultaneously in a quantum parallel computing manner, simulating various logical judgments and state transitions in the reasoning process. Compared with traditional Boolean logic-based reasoning simulations, quantum logic gate circuits can process more complex quantum state information and more comprehensively consider the uncertainty and multiple possibilities in the reasoning process, thereby more accurately evaluating the reasoning process of students' answers.

[0125] Beneficial effects: It can check the standardization and logic of the reasoning steps of students' answers. For example, in mathematical proof questions, it can accurately discover problems such as jumps in students' reasoning process and incorrect application of theorems. Through the efficient simulation of quantum logic gate circuits, it can quickly process the reasoning evaluation of a large number of students' answers and improve the efficiency of marking. And because of its good ability to handle complex reasoning situations, it can provide students with more detailed and accurate reasoning error feedback, help students better understand and improve their reasoning ability, and promote the improvement of students' logical thinking ability and in-depth mastery of subject knowledge.

[0126] The quantum data mining and machine learning algorithms in the multi-source data fusion insight module, taking the quantum clustering algorithm as an example, are similar to the concept of the quantum clustering algorithm in the previous intelligent review subsystem, but the application scenarios are different:

[0127] Suppose the data point set consisting of the writing features and learning behavior data to be clustered is X = {x1, x2, ..., x n}, where x i represents the i-th data point, which may be, for example, a vector composed of a student's writing pressure characteristic value, class note completeness characteristic value, etc.;

[0128] In quantum clustering algorithms, operations such as data points represented by quantum states and distance calculations based on quantum states may be involved, such as calculating the quantum distance between data points and:

[0129]

[0130] where |ψ(x i )> and |ψ(x j )>are the data points x i and x jThe corresponding quantum state representation, f is a function based on the quantum state, which is used to calculate the quantum distance, and may involve operations such as the inner product and norm of the quantum state and comparison operations of quantum bits.

[0131] Operation logic: Based on the quantum distance measurement between data points, similar writing features and learning behavior data are clustered into one category. By continuously adjusting the cluster center and re-dividing the data points, the similarity of data points in the same category in the quantum state space is as high as possible, and the similarity of data points between different categories is as low as possible. The quantum clustering algorithm utilizes the unique computing and storage characteristics of quantum states. For example, quantum bits can represent multiple states at the same time. In the process of calculating distances and updating cluster centers, multiple situations can be processed in parallel to improve clustering efficiency and effectiveness. In learning situation analysis, this clustering method can be used to mine different learning patterns of students from multi-source data. For example, the feature combinations of different types of students such as groups with learning difficulties and groups with excellent learning are discovered to provide data support for personalized teaching.

[0132] Beneficial effects: It can analyze the relationship between class notes and homework and test performance, and discover the learning difficulties or advantages corresponding to different writing and behavioral characteristics. For example, if it is found that a certain type of students writes slowly and has a high error rate in homework, they can be classified into one category through cluster analysis. Further analysis shows that it may be due to the lack of in-depth understanding of knowledge, which leads to long thinking time when writing and prone to errors in homework. Teachers can provide special knowledge counseling and learning method guidance for such students. The high efficiency of quantum clustering algorithms (using characteristics such as quantum parallel computing) can quickly process large-scale student multi-source data and provide teachers with learning situation analysis results in a timely manner, so that teachers can adjust teaching strategies in a timely manner, improve teaching effectiveness, and promote students' learning progress.

[0133] The quantum random walk algorithm in the psychological feature association prediction module is used to analyze the association between writing pressure, speed and other features and psychological factors in a high-dimensional feature space:

[0134] Similar to the quantum random walk algorithm in the previous knowledge graph reasoning and review module, but with different application scenarios and analysis objects, on a high-dimensional feature space graph G = (V, E) (where V is the vertex set and E is the edge set) constructed by writing features (such as pressure, speed, etc.) and psychological factors, the state of the quantum random walk can be represented by a quantum state at time, and its evolution equation can be roughly expressed as:

[0135] |ψ t+1 >=U|ψ t >

[0136] Among them, U is the unitary evolution operator of the quantum random walk, which is related to the structure of the high-dimensional feature space graph (the connection relationship between vertices) and the evolution rule of the quantum state. Its specific form can be expressed as (taking the simple discrete-time quantum random walk as an example):

[0137]

[0138] Here, A uv is the element of the adjacency matrix of the high-dimensional feature space graph, indicating whether there is an edge connecting vertex U and vertex v (if there is an edge, then A uv = 1; if there is no edge, then A uv = 0), |v><u| is the projection operator of the quantum state, and S is the local unitary transformation operator acting on the quantum state, such as the combination of basic quantum gate operations like Pauli matrices, etc., which is used to realize the transfer and state change of the quantum state between vertices. In this application scenario, vertices u and v can represent a certain combined state of writing eigenvalues or psychological factor eigenvalues. For example, vertex u may represent the state of medium writing pressure and low psychological pressure, and vertex v may represent the state of high writing pressure and high psychological pressure, etc. The edge represents a certain correlation or transfer possibility between these states.

[0139] Operating logic: Based on the structure of the high-dimensional feature space graph and the evolution rule of the quantum state, the quantum random walk starts from the initial state and, at each step, performs quantum state transfer and state change between the vertices of the combined state of writing features and psychological factors according to the unitary evolution operator. Through multiple iterations, the quantum state can explore multiple paths and state relationships simultaneously in a quantum parallel processing manner, and utilize the superposition and entanglement characteristics of the quantum state to挖掘出 high-dimensional feature space中深层次的关联信息,与传统的分析方法相比,量子随机游走能够更高效地遍历高维特征空间,发现更多复杂的书写特征与心理因素关联模式,因为它不受经典计算中串行处理的限制,能够同时考虑多种可能性,从而更准确地推断学生的心理状态变化,为教学过程中的心理干预提供更精准的依据。

[0140] Beneficial effects: It can infer the state changes such as psychological pressure and interest. For example, if it is found that during the quantum random walk process, when the writing speed gradually slows down and the stroke coherence becomes worse, the quantum state tends to the vertex of the state of high psychological pressure and low learning interest, then it can be inferred that the student may have relatively high psychological pressure and a decline in learning interest in this writing state. In this way, teachers can timely discover the psychological changes of students and take corresponding teaching strategy adjustments. Specific embodiment three:

[0142] As Figure 1-10 shown, based on the content in the above specific embodiment, the following content is further disclosed:

[0143] The operating steps of the above system architecture when in use are as follows;

[0144] Sp1: The writing feature collection and analysis subsystem operates first. The microscopic feature collection and analysis module uses highly sensitive quantum sensors to accurately capture microscopic dynamic data such as pen tip pressure, tilt angle and acceleration. Through quantum computing, a hand muscle movement model is constructed to analyze the stability of writing force, stroke continuity and emotional psychological state, and locate abnormal points. The macroscopic feature quantification and association module introduces a quantitative indicator system, uses quantum machine learning algorithms to quantify the aesthetic characteristics of fonts and associate them with psychological traits, and uses quantum Bayesian network algorithms to construct an association model between writing style and culture and subject preferences. At the same time, the algorithm parameters and prior probability determination are optimized to deeply explore hidden associations, laying the foundation for subsequent analysis;

[0145] Sp2: The intelligent review subsystem is immediately started. The semantic style fusion review module uses quantum natural language processing algorithms to understand the semantics of the answers, combines quantum neural networks to analyze writing style characteristics, and scores based on the correlation between the two. It assigns weights based on the font presentation and semantic accuracy of knowledge points, determines the knowledge mastery behind the font hesitation, and adjusts the scores. The knowledge graph reasoning review module uses quantum graph algorithms to construct subject knowledge graphs, locates logical associations in the graph based on student answers, checks the coverage and derivation process of knowledge nodes, simulates reasoning steps, and uses quantum search algorithms to evaluate open questions, provide score ranges and evaluation suggestions, and optimizes quantum neural network architecture and semantic vector processing to improve review efficiency and accuracy.

[0146] Sp3: The learning situation analysis and prediction subsystem starts working. The multi-source data fusion insight module builds a data integration platform, integrates writing characteristics and learning behavior data, uses quantum encryption to ensure data security, uses quantum hashing algorithm to preprocess data and uses quantum federated learning algorithm to explore correlation patterns, analyzes the learning status corresponding to writing and behavior characteristics, and accelerates analysis with the help of quantum computing; the psychological characteristic correlation prediction module builds a writing psychology analysis model, uses the quantum random walk algorithm to analyze the correlation between writing pressure, speed and other psychological factors, infers changes in psychological state, combines historical performance data, and uses the quantum autoregressive moving average model to build a performance prediction model, predicts future performance based on writing psychology and performance trends, warns and provides teaching suggestions, and adaptively adjusts algorithm parameters;

[0147] Sp4: The data visualization subsystem plays a role. It uses quantum rendering technology to integrate and process the data and results generated by the writing feature collection and analysis subsystem, the intelligent review subsystem, and the learning situation analysis and prediction subsystem. Through quantum three-dimensional visualization, it clearly displays the distribution of students' writing features in different dimensions and intuitively presents the relationship between them and their academic performance, greatly improving the clarity and interactivity of the visualization effect, allowing teachers, students and education managers to easily understand the complex information output by the system, so as to better carry out teaching, learning and management activities based on this information;

[0148] Sp5: The system self-optimization function is activated, and user feedback data and new learning sample data are collected regularly. The quantum reinforcement learning algorithm is used to deeply optimize the key parts of the system, such as the writing feature analysis model, the intelligent marking model, and the learning situation prediction model. According to the students' questioning information on the marking results and the new teaching syllabus requirements, the scoring rules and semantic understanding strategies in the intelligent marking model are automatically adjusted. At the same time, the parameters and structures of other models are adaptively modified, so that the system can closely fit the ever-changing education and teaching environment and the characteristics of the student group, continuously improve the overall performance and accuracy of the system, ensure that the system is always in an efficient operation state, and provide strong support for education and teaching. Specific embodiment three:

[0150] like Figure 1-10 As shown, according to the contents in the above specific embodiments, the following contents are further disclosed:

[0151] The specific process of achieving comprehensive and accurate intelligent review of answers in each module of the intelligent review subsystem is as follows:

[0152] Semantic style fusion review module

[0153] Semantic understanding: The quantum latent semantic analysis (QLSA) algorithm is used to deeply mine the semantic information in the answer text. QLSA can process large-scale text data and quickly identify the core concepts and themes in the answer and the semantic relationships between them through the parallel advantages of quantum computing. For example, in the answer to a subjective history question, it can accurately capture the historical events, figures, periods mentioned by students, and the causal relationship and influence relationship between them, so as to determine whether the answer fully and accurately responds to the requirements of the question at the semantic level;

[0154] Writing style analysis: quantum neural networks are used to encode and learn students' writing style characteristics. The unique architecture of quantum neural networks and the powerful computing power of quantum computing enable them to process complex writing style feature data, such as font size, shape, stroke thickness, inclination, writing speed change curve, etc. It can learn the potential patterns between different writing style characteristics and students' knowledge mastery and thinking state. For example, when students use larger fonts and bold strokes when writing important knowledge points, quantum neural networks can combine semantic information to judge that this may be the students' emphasis on this knowledge point, and if the semantics is correct, a certain positive evaluation weight will be given; on the contrary, if the font is small and the strokes are hesitant and there are many corrections, it may imply that the student is not confident in the knowledge or does not understand it deeply. Even if the semantics is roughly correct, the score will be adjusted accordingly;

[0155] Comprehensive evaluation: By organically integrating the results of semantic understanding with the results of writing style analysis, a comprehensive review is achieved. Instead of viewing semantics and style in isolation, the system comprehensively considers the impact of both on the quality of the answer based on pre-set fusion rules and model weights. For example, for a literary appreciation question, in addition to judging whether the student's understanding of the theme, character image, artistic techniques, etc. of the work is accurate (semantic level), the system will also analyze whether the writing style reflects the emotional resonance with the work, such as whether the fluency and rhythm of the font match the emotional expression of the work. If the semantics are accurate and the writing style can convey the emotions well, a high score will be given; if the semantics are biased and the writing style also shows that the student may not understand the work well (such as sloppy handwriting and chaotic structure), a lower score will be given, and detailed feedback suggestions can be generated to point out the semantic errors and problems reflected in the writing style, so as to achieve accurate and intelligent review.

[0156] Knowledge graph reasoning review module

[0157] In terms of knowledge graph construction and positioning: quantum graph algorithms are used to construct subject knowledge graphs. This algorithm can efficiently handle the organization and association of large-scale knowledge data. In the construction process, various concepts, principles, and knowledge points in the subject are used as nodes, and the logical relationships between them (such as causal relationships, inclusion relationships, derivation relationships, etc.) are used as edges to form a large and orderly knowledge network. When marking students' answers, the relevant concept networks and logical links are quickly located in the knowledge graph based on the knowledge points mentioned in the answers. For example, when marking a physical circuit analysis question, the system can quickly find nodes and their connection relationships related to circuit elements (resistors, capacitors, inductors, etc.) and circuit laws (Ohm's law, Kirchhoff's law, etc.) in the knowledge graph, so as to fully understand the knowledge scope and logical structure that students' answers should cover;

[0158] Reasoning process analysis: quantum logic gate circuits are used to simulate the reasoning process. This is based on the principle of quantum computing to accurately simulate and verify the logical reasoning steps. For math proof questions, logical reasoning questions, etc., the system converts the reasoning steps in the student's answer into an operation sequence of quantum logic gate circuits according to the reasoning rules and theorem system of the subject. For example, in a geometry proof question, each step of the deduction (such as drawing a new conclusion based on known conditions, using theorem to prove, etc.) corresponds to a specific operation combination in the quantum logic gate circuit. The system judges the correctness and rigor of the reasoning process by checking whether these operations conform to the rules of mathematical logic, whether there are logical jumps, circular reasoning, or incorrect application of theorems. For scientific experiment analysis questions, quantum logic gate circuits are also used to check the rationality of the experimental design, the accuracy of data processing, and the scientific nature of the conclusion deduction in the student's answer based on the experimental principles and the logical requirements of data analysis.

[0159] In terms of handling open-ended questions: quantum search algorithms are used to comprehensively explore and evaluate multiple possible answers to open-ended questions. Open-ended questions often do not have a single definite answer. Quantum search algorithms use the superposition and parallel characteristics of quantum computing to quickly traverse a large number of possible answer directions and ideas. For example, in a subjective question about the analysis of social phenomena, the quantum search algorithm can simultaneously consider different analysis angles, theoretical frameworks, and example references, and evaluate each possible answer path according to pre-set evaluation criteria (such as the depth, breadth, innovation, and rationality of the analysis). A reasonable score range is then derived, and inspiring evaluation suggestions can be provided to students, such as pointing out some unconsidered analysis angles or directions that can be further explored, thereby achieving intelligent review of open-ended questions.

[0160] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "comprising a reference structure" do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0161] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. Intelligent review and learning situation analysis system for subjective questions based on writing features, characterized by: The system architecture includes a writing feature collection and analysis subsystem, an intelligent review subsystem, and a learning situation analysis and prediction subsystem, among which: The writing feature collection and analysis subsystem is used to accurately collect multi-dimensional feature information in the student's writing process, and deeply analyze the student's writing habits, thinking patterns and psychological state information contained in these features. The writing feature collection and analysis subsystem includes a micro feature collection and analysis module and a macro feature quantification and association module; The intelligent review subsystem is used to review students' answers to subjective questions comprehensively, accurately and intelligently, comprehensively consider the semantic correctness of the answers, the characteristics of the writing style and the logical fit with the subject knowledge system, and give objective, fair and constructive review results and feedback. The intelligent review subsystem includes a semantic style fusion review module and an intelligent graph reasoning review module; The learning situation analysis and prediction subsystem: through deep integration of multi-source data, it explores the intrinsic connection and rules between writing characteristics and learning behavior, psychological state and academic performance, realizes all-round insight into students' learning situation, timely warning and accurate prediction, and provides a strong basis for personalized teaching and learning support. The learning situation analysis and prediction subsystem includes a multi-source data fusion insight module and a psychological characteristic association prediction module.

2. The system for intelligent reviewing and analyzing learning situations of subjective questions based on writing features according to claim 1 is characterized by: The micro-feature collection and analysis module uses a high-sensitivity quantum sensor to collect micro-dynamic data on pen tip pressure changes, tilt angles, and acceleration, and quickly builds a hand muscle movement model through quantum computing to analyze students' force stability, stroke continuity, and emotional state when writing, and accurately locate psychological fluctuations or knowledge uncertainty points reflected by abnormal changes in pressure and angles. The macro-feature quantification and association module introduces a multi-dimensional quantitative indicator system, uses a quantum machine learning algorithm to quantify the aesthetic features of fonts and analyze their association with psychological traits, and uses a quantum Bayesian network algorithm to build a writing style, cultural background, and subject preference association model to explore the influence of different cultures and subjects on the shaping of writing style and hidden associations.

3. The system for intelligent reviewing and analyzing learning situations of subjective questions based on writing features according to claim 1 is characterized by: The semantic style fusion review module adopts quantum natural language processing algorithm to understand the semantics of the answer, combines quantum neural network to analyze the writing style characteristics, comprehensively scores based on the connection between the two, gives reasonable weights based on the font emphasis and semantic accuracy when expressing knowledge points, judges the degree of mastery of knowledge points when the font is hesitant and adjusts the score. The knowledge graph reasoning review module uses quantum graph algorithm to construct subject knowledge graph, locates the associated logic in the graph based on the knowledge points of the students' answers, checks the coverage and derivation process of knowledge nodes, uses quantum logic gate circuits to simulate the reasoning process, disassembles the reasoning steps of mathematical proof questions, checks their standardization and logic, uses quantum search algorithm to evaluate the answer ideas of open questions, and provides score ranges and evaluation suggestions.

4. The system for intelligent reviewing and analyzing learning situations of subjective questions based on writing features according to claim 1 is characterized by: The multi-source data fusion insight module builds a data integration platform, integrates writing characteristics and learning behavior data, uses quantum encryption to ensure data security, explores the correlation pattern between the two through quantum data mining and machine learning algorithms, discovers the learning difficulties or advantages corresponding to different writing and behavior characteristics, and accelerates complex correlation analysis with the help of quantum computing. The psychological characteristic correlation prediction module constructs a writing psychology analysis model, uses the quantum random walk algorithm to analyze the characteristics of writing pressure and speed and the correlation with psychological factors in a high-dimensional feature space, infers the state changes of psychological pressure and interest, combines historical performance data, and uses quantum time series analysis and machine learning prediction algorithms to construct a performance prediction model. According to the writing psychological state and performance trend, the future performance trend is predicted, and warnings and teaching suggestions are provided for different situations to promote the coordinated development of students' physical and mental health and academic performance.

5. The system for intelligent reviewing and analyzing learning situations of subjective questions based on writing features according to claim 1 is characterized by: The system also includes a data visualization subsystem, which is used to present the data and results generated by the writing feature collection and analysis subsystem, the intelligent marking subsystem, and the learning situation analysis and prediction subsystem to teachers, students, and education administrators in an intuitive and visual manner. The subsystem uses quantum rendering technology to improve the clarity and interactivity of the visualization effect, so that users can more easily understand and use the information output by the system.

6. The system for intelligent reviewing and analyzing learning situations of subjective questions based on writing features according to claim 1 is characterized by: The system has a self-optimization function. By regularly collecting user feedback data and new learning sample data, it uses quantum reinforcement learning algorithms to adjust parameters and optimize structures of various models in the system, including writing feature analysis models, intelligent grading models, and learning situation prediction models, so as to adapt to the ever-changing educational and teaching environment and student group characteristics, and improve the performance and accuracy of the system.

7. The system for intelligent reviewing and analyzing learning situations of subjective questions based on writing features according to claim 1 is characterized by: When the macro-feature quantification and association module uses the quantum support vector machine to quantify the font aesthetic features, the kernel function adopts the quantum Gaussian kernel function, and its parameters are adjusted through the quantum optimization algorithm to improve the accuracy of the font aesthetic feature classification. When the association model is constructed using the quantum Bayesian network algorithm, the determination of the prior probability distribution is based on the quantum statistical analysis of large-scale sample data, so that the model can more accurately reflect the relationship between writing style and multiple factors.

8. The system for intelligent reviewing and analyzing learning situations of subjective questions based on writing features according to claim 1 is characterized by: The quantum neural network in the semantic-style fusion review module adopts a quantum convolutional neural network architecture. The convolution kernel size and step size of the quantum convolution layer are optimized according to the spatial distribution characteristics of the writing style features. When combining quantum latent semantic analysis to understand the semantics of the answer, the dimension of the semantic vector is reduced through a quantum principal component analysis algorithm to reduce the computational complexity and retain key semantic information, thereby improving the efficiency and accuracy of the comprehensive scoring.

9. The system for intelligent reviewing and analyzing learning situations of subjective questions based on writing features according to claim 1 is characterized by: In the process of constructing the subject knowledge graph, the knowledge graph reasoning and review module uses the quantum graph algorithm to traverse the knowledge nodes and discover relationships, so as to enhance the completeness and accuracy of the knowledge graph construction. When using quantum logic gate circuits to simulate the reasoning process, quantum error correction code technology is used to reduce the error rate in the quantum computing process, thereby ensuring the reliability of the reasoning step checks.

10. The system for intelligent reviewing and analyzing learning situations of subjective questions based on writing features according to claim 1, characterized in that: When fusing writing features and learning behavior data, the multi-source data fusion insight module uses a quantum hashing algorithm to pre-process the data to improve the speed and accuracy of data fusion, and when using quantum data mining and machine learning algorithms to explore association patterns, a quantum federated learning algorithm is used, so that when multi-source data are distributed in different storage locations, joint learning can still be effectively performed, protecting data privacy while mining valuable association information. When constructing a writing psychological analysis model, the psychological feature association prediction module uses a quantum random walk algorithm. The step size and number of iterations are adaptively adjusted according to the scale and complexity of the writing feature data to ensure efficient and accurate analysis of the association between features and psychological factors in a high-dimensional feature space. When constructing a performance prediction model, the quantum time series analysis algorithm uses a quantum autoregressive moving average model (QARIMA) to better capture the time series characteristics and trend changes of performance data.

Images