Intelligent analysis method for junior middle school English reading understanding test questions based on large model

Abstract

The invention discloses a junior high-school English reading understanding test question intelligent analysis method based on a large model, which comprises the following steps: S1, obtaining answer texts of junior high-school English reading understanding test questions completed by students and corresponding test question texts, and answers are from published standard answers; s2, preprocessing the input test question text and answer text; s3, performing semantic level deep analysis on the test question text by utilizing a deep context understanding large model; s4, according to the test question text and the standard answer, in combination with understanding of the logical reasoning large model on text context and logical relationship, analyzing the semantic matching degree of the student answer, and based on the standard answer and the logical reasoning large model, generating a semantic and logic comprehensive evaluation result of the student answer; s5, dynamically generating learning suggestions based on the semantic and logic comprehensive evaluation result of the student answers; and S6, outputting a test question analysis report. According to the invention, the answers of the students are analyzed more accurately, and targeted learning guidance is provided, so that the teaching effect and the learning experience of the students are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of educational technology, and in particular to an intelligent parsing method for English reading comprehension test questions for junior high school students based on a large model. Background Art

[0002] With the development of artificial intelligence technology, the field of education has gradually introduced intelligent tools to assist teaching and assessment. As an important part of English teaching, the test results of junior high school English reading comprehension can directly reflect students' reading ability, semantic understanding level and logical thinking ability. However, in the traditional teaching and assessment system, there are many challenges in the correction and feedback of English reading comprehension test questions.

[0003] At present, most schools and teaching assistance institutions still use manual marking to evaluate English reading comprehension test questions. Teachers need to review students' answers one by one, analyze their answer ideas against the standard answers, and give comments and scores. Although this can reflect the students' answering process, it has obvious limitations: due to the huge workload, it is difficult for teachers to efficiently mark a large number of test papers in a short period of time. In addition, teachers' subjective judgments inevitably lead to inconsistencies in scoring standards, affecting the objectivity of the evaluation.

[0004] In recent years, some automatic grading systems have begun to be applied in the field of education. They usually automatically score students' answers based on keyword matching and simple grammatical analysis. However, the automatic grading systems have exposed the following problems in actual applications: on the one hand, the keyword matching method cannot deeply understand the semantics of students' answers and easily ignores students' reasonable flexibility in expression; on the other hand, the automatic grading system has difficulty in handling complex logical relationships and lacks the ability to identify students' errors in the reasoning chain. In addition, the existing automatic grading technology lacks personalized feedback function and can only give simple correct and incorrect judgments, making it difficult to provide targeted learning suggestions for students' weaknesses.

[0005] To sum up, the existing technology has significant shortcomings in grading efficiency, evaluation accuracy and personalized feedback, and it is difficult to meet the needs of modern education scenarios for intelligent and efficient evaluation. These technical defects directly affect students' learning experience and teachers' teaching quality. A new method is urgently needed to solve the above problems. Summary of the invention

[0006] One purpose of the present invention is to propose a method for intelligent parsing of junior high school English reading comprehension test questions based on a large model. The present invention can more accurately parse students' answers and provide targeted learning guidance, thereby significantly improving teaching effectiveness and students' learning experience.

[0007] According to an embodiment of the present invention, a method for intelligently parsing junior high school English reading comprehension test questions based on a large model includes the following steps:

[0008] S1. Obtain the answer texts and corresponding test texts of junior high school English reading comprehension tests completed by students, where the answers are derived from published standard answers;

[0009] S2, preprocessing the input test question text and answer text, including formatting, grammar normalization and noise removal, to generate standardized text;

[0010] S3. Use the deep context understanding model to perform deep semantic analysis on the test text, analyze the contextual relationship, implicit semantic clues and logical structure in the test text, extract the main idea of ​​the article, the central idea of ​​the paragraph and the key sentences, perform semantic analysis on the student answer text, identify the logical reasoning process, answer ideas and implicit information in the language expression in the student answer, extract its relevance to the test text, perform multi-level context understanding based on the semantic analysis results, and generate semantic analysis results;

[0011] S4. Based on the test text and standard answers, combined with the understanding of the text context and logical relationship of the logical reasoning model, analyze the semantic matching degree of the student's answer, analyze the reasoning chain of the student's answer to identify logical errors or loopholes in the reasoning, and generate the semantic and logical comprehensive evaluation results of the student's answer based on the standard answer and the logical reasoning model, and mark the wrong parts in the answer;

[0012] S5. Dynamically generate learning suggestions based on the semantic and logical comprehensive evaluation results of students’ answers;

[0013] S6. Output the test question analysis report, which includes the standard answer, semantic analysis results, logical evaluation, student answer error marking and learning suggestions.

[0014] Optionally, S1 includes the following specific steps:

[0015] S11. Receive student's answer text A from a preset database or online system i And the corresponding test text Q i , where A i represents the answer text of the i-th student, Q i Indicates the reading comprehension test text corresponding to the answer;

[0016] S12, the received answer text A i and test text Q i Perform matching association to make the answer text A i With the test text Q i One-to-one correspondence, if the match fails, an exception prompt is returned:

[0017]

[0018] Among them, Match(A i ,Q i ) represents the matching result of the system between the answer text and the test text, θ is the matching threshold, when the similarity reaches or exceeds θ, the match is successful, otherwise an abnormal prompt is returned, Sim(A i ,Q i ) is to use cosine similarity to calculate the similarity of two texts:

[0019]

[0020] in, and is the word vector weight of the answer and question text, the weight of the kth word in the text;

[0021] S13, according to the preset format rules, the received answer text A i And test text Q i For storage, the identifier format is (A i ,Q i ).

[0022] Optionally, S2 includes the following specific steps:

[0023] S21, the received test text Q i and answer text A i Perform paragraph indentation, punctuation replacement and space deletion to obtain the formatted test text Q′ i and answer text A′ i ;

[0024] S22, formatted test text Q' i and answer text A′ i Perform grammatical standardization and automatically correct grammatical errors in the text, including corrections of inconsistent tenses, inconsistent subjects and predicates, and incorrect use of punctuation marks;

[0025] S23, formatted test text Q' i and answer text A′ i Perform denoising to remove meaningless symbols, redundant spaces and extra characters;

[0026] S24, the denoised test text Q′ i and answer text A′ i Standardize the text so that the vocabulary and expressions conform to the teaching standards, replace all non-standard words with standard forms, and obtain the standardized test text and answer text

[0027] S25. Verify the standardized test text and answer text If the format, syntax or content does not meet the standard, an exception prompt will be returned and marked as Otherwise, the standardized test text and answer text Pass to the next processing step.

[0028] Optionally, S3 includes the following specific steps:

[0029] S31. Using the deep context understanding model to analyze the standardized test text Perform deep analysis of the semantic level and generate multi-level semantic representations, including word-level embedding vectors Sentence-level representation vector and the paragraph-level representation vector

[0030] S32. Analyze the test text The contextual relationship, implicit semantic clues and logical structure in the text are used to extract the main idea of ​​the article. i 、C. Paragraph main idea ij and key sentence K ij ;

[0031] Article Subject i By maximizing the similarity between the chapter-level representation vector and the topic vector T, we can obtain:

[0032]

[0033] Among them, Sim(·,·) is the similarity calculation function, and T is the predefined topic vector set;

[0034] Paragraph main idea C ij It is obtained by clustering the sentence-level representation vectors within the paragraph;

[0035] Key sentence K ij By calculating the sentence-level representation vector and the article theme M i Determination of relevance:

[0036]

[0037] Among them, δ is the correlation threshold.

[0038] S33. Standardized student answer text Perform semantic analysis to generate semantic representation vectors of student answers Calculate word-level embeddings using the same method as the test text Sentence-level representation and paragraph-level representation

[0039] S34, identify the logical reasoning process, answer ideas and implicit information in the language expression in the student's answer, extract its relevance to the test text, and calculate the semantic matching degree S between the student's answer and the test text i :

[0040]

[0041] S35. Based on the standardized test text and standard answers, use the logical reasoning model to generate a standard reasoning chain diagram:

[0042]

[0043] in, is the set of concept nodes in the standard answer, is the set of logical relationship edges between concepts;

[0044] Construct a reasoning chain diagram of the student's answer:

[0045]

[0046] in, is the set of concept nodes in the student’s answer, is the set of logical relationship edges between nodes;

[0047] S36, Comparison of reasoning chains of students' test questions With the standard reasoning chain diagram Evaluate the completeness and correctness of students' reasoning process and calculate the structural similarity of the reasoning chain:

[0048]

[0049] in, is the similarity of the reasoning chain graph structure between the student answer and the standard answer, Represents the set of matching relationship nodes between two graphs, Represents the total number of nodes in the standard reasoning chain graph;

[0050] S37. Based on the results of semantic analysis, and student answer text Perform multi-level contextual understanding, identify irony, metaphor or metaphor rhetoric in the test questions, and generate semantic analysis results, including the degree of semantic and logical match between the student's answer and the test text, as well as annotations of comprehension deviations and reasoning errors in the student's answer:

[0051]

[0052] Among them, R iIt represents the semantic analysis result of the i-th test question, which is a multi-dimensional comprehensive evaluation result of the student's answer text and the test question text. Indicates the matching degree between the rhetoric recognition result in the test text and the total rhetoric set, represents the set of rhetorical devices identified in the i-th test text, R total represents the total set of predefined rhetorical devices, λ1,λ2,λ3 represent weighted coefficients of different dimensions, which are used to balance the importance of semantic matching, logical reasoning, reasoning chain matching and rhetoric recognition in the parsing results.

[0053] Optionally, the word-level embedding vector Indicates the test text The embedding of the jth word in , the sentence-level representation vector is obtained by weighted averaging the word-level embeddings:

[0054]

[0055] Among them, N k is the number of words in the kth sentence, is the attention weight of the jth word, calculated as:

[0056]

[0057] Among them, u is the context global semantic vector;

[0058] Chapter-level representation vector Obtained by performing encoder processing on the sentence-level representation vector.

[0059] Optionally, S5 includes the following specific steps:

[0060] S51. According to the comprehensive evaluation results R i Identify students’ weaknesses in semantic understanding and logical reasoning and generate a set of weaknesses:

[0061] B i = {b ij ∣Error ij =|S ij -L ij |≥∈};

[0062] Among them, Error ij Indicates that students have a semantic matching degree S ij Similarity with logic L ij The deviation between them, ∈ is the error threshold, and the aspects with errors greater than the threshold are marked as weak points b ij ;

[0063] S52. Retrieve the students’ weak points from the question bank ijRelated recommended test questions collection T ij , select appropriate test questions by matching the context similarity and the reasoning chain graph structure:

[0064]

[0065] in, is the set of test questions in the question bank, α1 and β1 are the weight coefficients of semantic and logical matching;

[0066] S53, based on students’ learning trajectory and semantic matching S i Generate personalized learning path P i :

[0067]

[0068] in, Represents finding the path P that makes the learning path best i ,ω j Weak point b ij The weight coefficient in path optimization indicates the importance of the weak point in the learning path. For weak point b ij The average quality rating of the matched test set, For semantic similarity, calculate the question vector V t Vulnerability semantic vector The similarity between Represents the recommended test question reasoning chain graph G t Reasoning chain diagram with standard test questions The structural similarity between them, ∥G t ∥ represents the module length of the test question reasoning chain diagram, α2 and β2 are the weighting coefficients of semantic similarity and logical reasoning similarity.

[0069] Optionally, S6 includes the following specific steps:

[0070] S61. Generate a test question analysis report that includes a comparison between the standard answer and the student's answer. The test question analysis report shows the difference between the student's answer and the standard answer, and marks each error to highlight the misunderstanding and logical errors in the student's answering process;

[0071] S62, semantic analysis section describes the students' performance in understanding the main idea of ​​the text, the central idea of ​​the paragraph and the key sentences. The test question analysis report highlights the students' scores and weaknesses at the semantic level and indicates the understanding links that need to be strengthened;

[0072] S63, the logic assessment part analyzes whether the student's reasoning chain is complete, determines whether there are jumps and unreasonable inferences in logical thinking, and presents the specific links where the logic problems are in the test question analysis report;

[0073] S64. Based on the results of semantic analysis and logical assessment, the test analysis report provides specific improvement directions and learning methods for students' weak points in semantic understanding and logical reasoning, and recommends exercises related to the weak points.

[0074] The beneficial effects of the present invention are:

[0075] (1) The present invention adopts a multi-level semantic embedding representation, constructs text representation from the word level, sentence level and paragraph level respectively, and generates a reasoning chain diagram in combination with a large logical reasoning model. In the semantic matching process, not only the similarity calculation of word vectors and sentence-level embeddings is used to improve the matching degree between the answer and the test text, but also the similarity of the reasoning chain structure is used to evaluate the logical process of students in answering questions. The traditional keyword matching method can only capture shallow semantics. The algorithm of the present invention significantly improves the parsing accuracy and logical reasoning ability of complex text semantics through the combination of deep semantic analysis and logical reasoning, ensuring a more comprehensive understanding of students' answers.

[0076] (2) The present invention combines the calculation results of semantic matching degree with the similarity of reasoning chain structure, and dynamically generates the optimal learning path according to the students' performance in different weak points. It not only considers the students' current knowledge mastery, but also dynamically sorts and adjusts the difficulty of the exercises in the learning path based on the similarity of the semantics and logical structure of the recommended test questions, ensuring that each student can obtain learning resources that match their knowledge level, avoiding the learning frustration caused by inappropriate difficulty of exercises in traditional methods, and achieving a more efficient learning experience.

[0077] (3) The present invention introduces three modules, namely semantic analysis, logical evaluation and personalized learning suggestions, in the generation of test analysis reports. Each incorrect item is accurately marked through a multi-dimensional comparison between the standard answer and the student's answer, and the analysis results at the semantic level and the logical level are presented in detail in the report. Different from the prior art that only provides a simple judgment on whether the answer is correct or not, the analysis report of the present invention can not only point out the specific weaknesses of the students, but also generate targeted learning suggestions based on the reasoning chain analysis, further enhancing the students' room for improvement in semantic understanding and logical reasoning. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0079] Figure 1 A flowchart of a method for intelligently parsing English reading comprehension questions for junior high school students based on a large model proposed by the present invention;

[0080] Figure 2This is a schematic diagram of the personalized learning path generation process in the intelligent parsing method for junior high school English reading comprehension test questions based on a large model proposed by the present invention. DETAILED DESCRIPTION

[0081] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0082] refer to Figure 1-Figure 2 , a method for intelligent parsing of junior high school English reading comprehension test questions based on a large model, comprising the following steps:

[0083] S1. Obtain the answer texts and corresponding test texts of junior high school English reading comprehension tests completed by students, where the answers are derived from published standard answers;

[0084] S2, preprocessing the input test question text and answer text, including formatting, grammar normalization and noise removal, to generate standardized text;

[0085] S3. Use the deep context understanding model to perform deep semantic analysis on the test text, analyze the contextual relationship, implicit semantic clues and logical structure in the test text, extract the main idea of ​​the article, the central idea of ​​the paragraph and the key sentences, perform semantic analysis on the student answer text, identify the logical reasoning process, answer ideas and implicit information in the language expression in the student answer, extract its relevance to the test text, perform multi-level context understanding based on the semantic analysis results, and generate semantic analysis results;

[0086] S4. Based on the test text and standard answers, combined with the understanding of the text context and logical relationship of the logical reasoning model, analyze the semantic matching degree of the student's answer, analyze the reasoning chain of the student's answer to identify logical errors or loopholes in the reasoning, and generate the semantic and logical comprehensive evaluation results of the student's answer based on the standard answer and the logical reasoning model, and mark the wrong parts in the answer;

[0087] S5. Dynamically generate learning suggestions based on the semantic and logical comprehensive evaluation results of students’ answers;

[0088] S6. Output the test question analysis report, which includes the standard answer, semantic analysis results, logical evaluation, student answer error marking and learning suggestions.

[0089] In this implementation, S1 includes the following specific steps:

[0090] S11. Receive student's answer text A from a preset database or online system i And the corresponding test text Q i , where Ai represents the answer text of the i-th student, Q i Indicates the reading comprehension test text corresponding to the answer;

[0091] S12, the received answer text A i and test text Q i Perform matching association to make the answer text A i With the test text Q i One-to-one correspondence, if the match fails, an exception prompt is returned:

[0092]

[0093] Among them, Match(A i ,Q i ) represents the matching result of the system between the answer text and the test text, θ is the matching threshold, when the similarity reaches or exceeds θ, the match is successful, otherwise an abnormal prompt is returned, Sim(A i ,Q i ) is to use cosine similarity to calculate the similarity of two texts:

[0094]

[0095] in, and is the word vector weight of the answer and question text, the weight of the kth word in the text;

[0096] S13, according to the preset format rules, the received answer text A i And test text Q i For storage, the identifier format is (A i ,Q i ).

[0097] In this implementation, S2 includes the following specific steps:

[0098] S21, the received test text Q i and answer text A i Perform paragraph indentation, punctuation replacement and space deletion to obtain the formatted test text Q′ i and answer text A′ i ;

[0099] S22, formatted test text Q' i and answer text A′ i Perform grammatical standardization and automatically correct grammatical errors in the text, including corrections of inconsistent tenses, inconsistent subjects and predicates, and incorrect use of punctuation marks;

[0100] S23, formatted test text Q'i and answer text A′ i Perform denoising to remove meaningless symbols, redundant spaces and extra characters;

[0101] S24, the denoised test text Q′ i and answer text A′ i Standardize the text so that the vocabulary and expressions conform to the teaching standards, replace all non-standard words with standard forms, and obtain the standardized test text and answer text

[0102] S25. Verify the standardized test text and answer text If the format, syntax or content does not meet the standard, an exception prompt will be returned and marked as Otherwise, the standardized test text and answer text Pass to the next processing step.

[0103] In this implementation, S3 includes the following specific steps:

[0104] S31. Using the deep context understanding model to analyze the standardized test text Perform deep analysis of the semantic level and generate multi-level semantic representations, including word-level embedding vectors Sentence-level representation vector and the paragraph-level representation vector

[0105] S32. Analyze the test text The contextual relationship, implicit semantic clues and logical structure in the text are used to extract the main idea of ​​the article. i 、C. Paragraph main idea ij and key sentence K ij ;

[0106] Article Subject i By maximizing the similarity between the chapter-level representation vector and the topic vector T, we can obtain:

[0107]

[0108] Among them, Sim(·,·) is the similarity calculation function, and T is the predefined topic vector set;

[0109] Paragraph main idea C ij It is obtained by clustering the sentence-level representation vectors within the paragraph;

[0110] Key sentence K ijBy calculating the sentence-level representation vector and the article theme M i Determination of relevance:

[0111]

[0112] Among them, δ is the correlation threshold.

[0113] S33. Standardized student answer text Perform semantic analysis to generate semantic representation vectors of student answers Calculate word-level embeddings using the same method as the test text Sentence-level representation and paragraph-level representation

[0114] S34, identify the logical reasoning process, answer ideas and implicit information in the language expression in the student's answer, extract its relevance to the test text, and calculate the semantic matching degree S between the student's answer and the test text i :

[0115]

[0116] S35. Based on the standardized test text and standard answers, use the logical reasoning model to generate a standard reasoning chain diagram:

[0117]

[0118] in, is the set of concept nodes in the standard answer, is the set of logical relationship edges between concepts;

[0119] Construct a reasoning chain diagram of the student's answer:

[0120]

[0121] in, is the set of concept nodes in the student’s answer, is the set of logical relationship edges between nodes;

[0122] S36, Comparison of reasoning chains of students' test questions With the standard reasoning chain diagram Evaluate the completeness and correctness of students' reasoning process and calculate the structural similarity of the reasoning chain:

[0123]

[0124] in, is the similarity of the reasoning chain graph structure between the student answer and the standard answer, Represents the set of matching relationship nodes between two graphs, Represents the total number of nodes in the standard reasoning chain graph;

[0125] S37. Based on the results of semantic analysis, and student answer text Perform multi-level contextual understanding, identify irony, metaphor or metaphor rhetoric in the test questions, and generate semantic analysis results, including the degree of semantic and logical match between the student's answer and the test text, as well as annotations of comprehension deviations and reasoning errors in the student's answer:

[0126]

[0127] Among them, R i It represents the semantic analysis result of the i-th test question, which is a multi-dimensional comprehensive evaluation result of the student's answer text and the test question text. Indicates the matching degree between the rhetoric recognition result in the test text and the total rhetoric set, represents the set of rhetorical devices identified in the i-th test text, R total represents the total set of predefined rhetorical devices, λ1,λ2,λ3 represent weighted coefficients of different dimensions, which are used to balance the importance of semantic matching, logical reasoning, reasoning chain matching and rhetoric recognition in the parsing results.

[0128] In this implementation, the word-level embedding vector Indicates the test text The embedding of the jth word in , the sentence-level representation vector is obtained by weighted averaging the word-level embeddings:

[0129]

[0130] Among them, N k is the number of words in the kth sentence, is the attention weight of the jth word, calculated as:

[0131]

[0132] Among them, u is the context global semantic vector;

[0133] Chapter-level representation vector Obtained by performing encoder processing on the sentence-level representation vector.

[0134] In this implementation, S5 includes the following specific steps:

[0135] S51. According to the comprehensive evaluation results R i Identify students’ weaknesses in semantic understanding and logical reasoning and generate a set of weaknesses:

[0136] B i= {b ij |Error ij =|S ij -L ij |≥∈};

[0137] Among them, Error ij Indicates that students have a semantic matching degree S ij Similarity with logic L ij The deviation between them, ∈ is the error threshold, and the aspects with errors greater than the threshold are marked as weak points b ij ;

[0138] S52. Retrieve the students’ weak points from the question bank ij Related recommended test questions collection T ij , select appropriate test questions by matching the context similarity and the reasoning chain graph structure:

[0139]

[0140] in, is the set of test questions in the question bank, α1 and β1 are the weight coefficients of semantic and logical matching;

[0141] S53, based on students’ learning trajectory and semantic matching S i Generate personalized learning path P i :

[0142]

[0143] in, Represents finding the path P that makes the learning path best i ,ω j Weak point b ij The weight coefficient in path optimization indicates the importance of the weak point in the learning path. For weak point b ij The average quality rating of the matched test set, For semantic similarity, calculate the question vector V t Vulnerability semantic vector The similarity between Represents the recommended test question reasoning chain graph G t Reasoning chain diagram with standard test questions The structural similarity between them, ∥G t ∥ represents the module length of the test question reasoning chain diagram, α2 and β2 are the weighting coefficients of semantic similarity and logical reasoning similarity.

[0144] In this implementation, S6 includes the following specific steps:

[0145] S61. Generate a test question analysis report that includes a comparison between the standard answer and the student's answer. The test question analysis report shows the difference between the student's answer and the standard answer, and marks each error to highlight the misunderstanding and logical errors in the student's answering process;

[0146] S62, semantic analysis section describes the students' performance in understanding the main idea of ​​the text, the central idea of ​​the paragraph and the key sentences. The test question analysis report highlights the students' scores and weaknesses at the semantic level and indicates the understanding links that need to be strengthened;

[0147] S63, the logic assessment part analyzes whether the student's reasoning chain is complete, determines whether there are jumps and unreasonable inferences in logical thinking, and presents the specific links where the logic problems are in the test question analysis report;

[0148] S64. Based on the results of semantic analysis and logical assessment, the test analysis report provides specific improvement directions and learning methods for students' weak points in semantic understanding and logical reasoning, and recommends exercises related to the weak points.

[0149] Embodiment 1:

[0150] In March 2024, a key middle school in a certain city adopted the intelligent analysis system of the present invention in the final examination of the second grade of junior high school. It was responsible for correcting the English reading comprehension questions of 250 students. Each test paper contained 5 reading comprehension articles with a total of 25 questions. The question types covered detailed understanding, reasoning and judgment, word meaning guessing and paragraph summary. Due to the huge amount of correction and the need for accurate personalized feedback, the school decided to use this system for intelligent correction to verify its actual effect.

[0151] Teacher Wang uploaded the scanned copies of all 250 test papers to the intelligent parsing system starting at 14:30 on March 20, 2024. The system automatically read the text and formatted the test papers and answer texts.

[0152] The answer of one of the students caught the attention of the system. The student's answer to the seven-choice five question was as follows:

[0153] Questions:

[0154] You have the chance to stay at home by yourself! You may feelhappy.But you could also feel a bit worried 1._______.Don't be afraid you cando things to get away from danger when you are at home alone and also learn how to deal with an emergency if it happens.

[0155] Student answer: E. because you want to be independent;

[0156] Intelligent parsing process:

[0157] Time: The system started parsing the student's test questions at 14:37.

[0158] Semantic analysis: The system first identifies the key point of the question as “But you could also feel a bit worried,” and determines based on the context that the sentence needs to explain the reason for the worry rather than express an independent wish.

[0159] Logical reasoning: The system generates a chain of reasoning through a logical reasoning model and finds that the semantics of the student's answer does not match the contextual logic. The system automatically marks the error and points out in the report: "The student mistakenly understands the reason for his worry as a desire for independence, but according to the context of the article, this should describe the difficulties that the student may face when he is alone at home."

[0160] The system suggests the correct answer is:

[0161] F.because you may face with some difficulties that you never met before;

[0162] Personalized feedback and learning path generation:

[0163] The system generated a detailed personalized feedback report at 15:10, which pointed out the student's weaknesses in reasoning and judgment questions, and recommended 5 related practice questions. The questions were all from the school question bank and were highly similar to the questions that the student answered incorrectly in terms of semantics and logical structure. The system also suggested that the student read the article "How to Be Safe When Home Alone" to further consolidate this type of semantic inference ability.

[0164] Marking time: 250 papers must be marked before 15:45, with a total time of 1 hour and 15 minutes.

[0165] Evaluation consistency: The consistency of the intelligent system in semantic matching and logical inference reached 98.5%.

[0166] Teacher feedback: Teacher Wang said that the system grading greatly reduced the workload and saved nearly 80% of the time than expected.

[0167] Student feedback: After receiving the feedback report, 73% of the students had a clear understanding of their misunderstandings and said that their weak points were effectively improved through the recommended exercises.

[0168] On April 10, 2024, the school conducted another mock exam to assess students’ progress on reasoning and word meaning guessing questions. The data showed that the class’s overall accuracy rate for this type of question increased from 78% to 89%, with the students with the highest error rate increasing their accuracy rate by 18 percentage points.

[0169] Through the verification of this example, the efficiency and accuracy of the intelligent analysis system in large-scale examination paper grading have been fully demonstrated. The system can not only quickly complete the grading task, but also provide detailed personalized feedback to help students learn and improve their weaknesses. At the same time, the high consistency of the system ensures the objectivity of the evaluation results, significantly improving the work efficiency and teaching effectiveness of teachers.

[0170] The present invention adopts multi-level semantic embedding representation, constructs text representation from word level, sentence level and paragraph level respectively, and generates a reasoning chain diagram in combination with a large logical reasoning model. In the semantic matching process, not only the matching degree between the answer and the test text is improved by similarity calculation of word vector and sentence-level embedding, but also the logical process of students in answering questions is evaluated by the similarity of reasoning chain structure. The traditional keyword matching method can only capture shallow semantics. The algorithm of the present invention significantly improves the parsing accuracy and logical reasoning ability of complex text semantics through the combination of deep semantic analysis and logical reasoning, ensuring a more comprehensive understanding of students' answers.

[0171] The present invention combines the calculation results of semantic matching degree with the similarity of reasoning chain structure, and dynamically generates the optimal learning path according to the students' performance in different weak points. It not only takes into account the students' current knowledge mastery, but also dynamically sorts and adjusts the difficulty of the exercises in the learning path based on the similarity of the semantics and logical structures of the recommended test questions, ensuring that each student can obtain learning resources that match their knowledge level, avoiding the learning frustration caused by inappropriate difficulty of exercises in traditional methods, and achieving a more efficient learning experience.

[0172] The present invention introduces three modules of semantic analysis, logical evaluation and personalized learning suggestions in the generation of test question analysis reports. Each incorrect item is accurately marked through a multi-dimensional comparison between standard answers and student answers, and the analysis results at the semantic level and logical level are presented in detail in the report. Different from the prior art that only provides a simple judgment on whether the answer is correct or not, the analysis report of the present invention can not only point out the students' specific weaknesses but also generate targeted learning suggestions based on reasoning chain analysis, further enhancing students' room for improvement in semantic understanding and logical reasoning.

[0173] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for intelligent analysis of junior high school English reading comprehension test questions based on a large model, characterized in that: The steps include: S1. Obtain the answer texts and corresponding test texts of junior high school English reading comprehension tests completed by students, where the answers are derived from published standard answers; S2, preprocessing the input test question text and answer text, including formatting, grammar normalization and noise removal, to generate standardized text; S3. Use the deep context understanding model to perform deep semantic analysis on the test text, analyze the contextual relationship, implicit semantic clues and logical structure in the test text, extract the main idea of ​​the article, the central idea of ​​the paragraph and the key sentences, perform semantic analysis on the student answer text, identify the logical reasoning process, answer ideas and implicit information in the language expression in the student answer, extract its relevance to the test text, perform multi-level context understanding based on the semantic analysis results, and generate semantic analysis results; S4. Based on the test text and standard answers, combined with the understanding of the text context and logical relationship of the logical reasoning model, analyze the semantic matching degree of the student's answer, analyze the reasoning chain of the student's answer to identify logical errors or loopholes in the reasoning, and generate the semantic and logical comprehensive evaluation results of the student's answer based on the standard answer and the logical reasoning model, and mark the wrong parts in the answer; S5. Dynamically generate learning suggestions based on the semantic and logical comprehensive evaluation results of students’ answers; S6. Output the test question analysis report, which includes the standard answer, semantic analysis results, logical evaluation, student answer error marking and learning suggestions.

2. According to claim 1, a method for intelligent analysis of junior high school English reading comprehension test questions based on a large model is characterized in that: The S1 comprises the following specific steps: S11. Receive student's answer text A from a preset database or online system i And the corresponding test text Q i , where A i represents the answer text of the i-th student, Q i Indicates the reading comprehension test text corresponding to the answer; S12, the received answer text A i and test text Q i Perform matching association to make the answer text A i With the test text Q i One-to-one correspondence, if the match fails, an exception prompt is returned: Among them, Match(A i ,Q i ) represents the matching result of the system between the answer text and the test text, θ is the matching threshold, when the similarity reaches or exceeds θ, the match is successful, otherwise an abnormal prompt is returned, Sim(A i ,Q i ) is to use cosine similarity to calculate the similarity of two texts: in, and is the word vector weight of the answer and question text, the weight of the kth word in the text; S13, according to the preset format rules, the received answer text A i And test text Q i For storage, the identifier format is (A i ,Q i ).

3. The intelligent analysis method for junior high school English reading comprehension test questions based on a large model according to claim 1 is characterized in that: The S2 comprises the following specific steps: S21, the received test text Q i and answer text A i Perform paragraph indentation, punctuation replacement and space deletion to obtain the formatted test text Q′ i and answer text A′ i ; S22, formatted test text Q' i and answer text A′ i Perform grammatical standardization and automatically correct grammatical errors in the text, including corrections of inconsistent tenses, inconsistent subjects and predicates, and incorrect use of punctuation marks; S23, formatted test text Q' i and answer text A′ i Perform denoising to remove meaningless symbols, redundant spaces and extra characters; S24, the denoised test text Q′ i and answer text A′ i Standardize the text so that the vocabulary and expressions conform to the teaching standards, replace all non-standard words with standard forms, and obtain the standardized test text and answer text S25. Verify the standardized test text and answer text If the format, syntax or content does not meet the standard, an exception prompt will be returned and marked as Otherwise, the standardized test text and answer text Pass to the next processing step.

4. The intelligent analysis method for junior high school English reading comprehension test questions based on a large model according to claim 1 is characterized in that: The S3 includes the following specific steps: S31. Using the deep context understanding model to analyze the standardized test text Perform deep analysis of the semantic level and generate multi-level semantic representations, including word-level embedding vectors Sentence-level representation vector and the paragraph-level representation vector S32. Analyze the test text The contextual relationship, implicit semantic clues and logical structure in the text are used to extract the main idea of ​​the article. i 、C. Paragraph main idea ij and key sentence K ij ; Article Subject i By maximizing the similarity between the chapter-level representation vector and the topic vector T, we can obtain: Among them, Sim(·,·) is the similarity calculation function, and T is the predefined topic vector set; Paragraph main idea C ij It is obtained by clustering the sentence-level representation vectors within the paragraph; Key sentence K ij By calculating the sentence-level representation vector and the article theme M i Determination of relevance: Among them, δ is the correlation threshold. S33. Standardized student answer text Perform semantic analysis to generate semantic representation vectors of student answers Calculate word-level embeddings using the same method as the test text Sentence-level representation and paragraph-level representation S34, identify the logical reasoning process, answer ideas and implicit information in the language expression in the student's answer, extract its relevance to the test text, and calculate the semantic matching degree S between the student's answer and the test text i : S35. Based on the standardized test text and standard answers, use the logical reasoning model to generate a standard reasoning chain diagram: in, is the set of concept nodes in the standard answer, is the set of logical relationship edges between concepts; Construct a reasoning chain diagram of the student's answer: in, is the set of concept nodes in the student’s answer, is the set of logical relationship edges between nodes; S36, Comparison of reasoning chains of students' test questions With the standard reasoning chain diagram Evaluate the completeness and correctness of students' reasoning process and calculate the structural similarity of the reasoning chain: in, is the similarity of the reasoning chain graph structure between the student answer and the standard answer, Represents the set of matching relationship nodes between two graphs, Represents the total number of nodes in the standard reasoning chain graph; S37. Based on the results of semantic analysis, and student answer text Perform multi-level contextual understanding, identify irony, metaphor or metaphor rhetoric in the test questions, and generate semantic analysis results, including the degree of semantic and logical match between the student's answer and the test text, as well as annotations of comprehension deviations and reasoning errors in the student's answer: Among them, R i It represents the semantic analysis result of the i-th test question, which is a multi-dimensional comprehensive evaluation result of the student's answer text and the test question text. Indicates the matching degree between the rhetoric recognition result in the test text and the total rhetoric set, represents the set of rhetorical devices identified in the i-th test text, R total represents the total set of predefined rhetorical devices, λ1,λ2,λ3 represent weighted coefficients of different dimensions, which are used to balance the importance of semantic matching, logical reasoning, reasoning chain matching and rhetoric recognition in the parsing results.

5. The intelligent analysis method for junior high school English reading comprehension test questions based on a large model according to claim 4 is characterized in that: The word-level embedding vector Indicates the test text The embedding of the jth word in , the sentence-level representation vector is obtained by weighted averaging the word-level embeddings: Among them, N k is the number of words in the kth sentence, is the attention weight of the jth word, calculated as: Among them, u is the context global semantic vector; Chapter-level representation vector Obtained by performing encoder processing on the sentence-level representation vector.

6. The intelligent analysis method for junior high school English reading comprehension test questions based on a large model according to claim 1 is characterized in that: The S5 comprises the following specific steps: S51. According to the comprehensive evaluation results R i Identify students’ weaknesses in semantic understanding and logical reasoning and generate a set of weaknesses: B i ={b ij ∣Error ij =|S ij -L ij |≥∈}; Among them, Error ij Indicates that students have a semantic matching degree S ij Similarity with logic L ij The deviation between them, ∈ is the error threshold, and the aspects with errors greater than the threshold are marked as weak points b ij ; S52. Retrieve the students’ weak points from the question bank ij Related recommended test questions collection T ij , select appropriate test questions by matching the context similarity and the reasoning chain graph structure: in, is the set of test questions in the question bank, α1 and β1 are the weight coefficients of semantic and logical matching; S53, based on students’ learning trajectory and semantic matching S i Generate personalized learning path P i : in, Represents finding the path P that makes the learning path best i ,ω j Weak point b ij The weight coefficient in path optimization indicates the importance of the weak point in the learning path. For weak point b ij The average quality rating of the matched test set, For semantic similarity, calculate the question vector V t Vulnerability semantic vector The similarity between Represents the recommended test question reasoning chain graph G t Reasoning chain diagram with standard test questions The structural similarity between them, ∥G t ∥ represents the module length of the test question reasoning chain diagram, α2 and β2 are the weighting coefficients of semantic similarity and logical reasoning similarity.

7. The intelligent analysis method for junior high school English reading comprehension test questions based on a large model according to claim 1 is characterized in that: The S6 comprises the following specific steps: S61. Generate a test question analysis report that includes a comparison between the standard answer and the student's answer. The test question analysis report shows the difference between the student's answer and the standard answer, and marks each error to highlight the misunderstanding and logical errors in the student's answering process; S62, semantic analysis section describes the students' performance in understanding the main idea of ​​the text, the central idea of ​​the paragraph and the key sentences. The test question analysis report highlights the students' scores and weaknesses at the semantic level and indicates the understanding links that need to be strengthened; S63, the logic assessment part analyzes whether the student's reasoning chain is complete, determines whether there are jumps and unreasonable inferences in logical thinking, and presents the specific links where the logic problems are in the test question analysis report; S64. Based on the results of semantic analysis and logical assessment, the test analysis report provides specific improvement directions and learning methods for students' weak points in semantic understanding and logical reasoning, and recommends exercises related to the weak points.

Images