Intelligent rewriting method for reading and understanding proposition materials based on large model

Abstract

The invention discloses an intelligent rewriting method for reading and understanding proposition materials based on a large model. The method comprises the steps that S1, an English original text input by a user is received; s2, preprocessing the English original text; s3, automatically identifying and extracting a core topic representation model and a context information set of the original English text based on the constructed text representation model by utilizing a pre-trained GPT large model; s4, generating a text feature set containing vocabulary difficulty, sentence pattern complexity and grammar complexity; s5, the GPT large model carries out self-adaptive simplification on the original English text according to the junior and middle school English teaching outline and the proposition standard of the middle school examination; s6, performing grammar correctness, semantic integrity and logic continuity evaluation on the English original text rewriting version generated after simplification by the GPT large model; s7, adaptively adjusting a simplified standard and a generation strategy; and S8, outputting the English text which is simplified by the GPT large model and meets the proposition standard of the Chinese examination to the user. The teaching efficiency is improved, and the quality and diversity of the generated content are ensured.

Description

Technical Field

[0001] The present invention relates to the field of English technology, and in particular to an intelligent rewriting method for reading comprehension proposition materials based on a large model. Background Art

[0002] With the rapid development of artificial intelligence and natural language processing technologies, the application of intelligent tools in the field of education has gradually become popular. In English teaching, reading comprehension, as an important part of cultivating students' language application and thinking abilities, has become the focus of attention of teachers and students. However, in the traditional teaching process, the proposition, rewriting and evaluation of English reading comprehension mainly rely on teachers' manual operations, which is not only inefficient but also has many limitations.

[0003] In the prior art, the proposition and material rewriting of English reading comprehension are usually completed manually by teachers or teaching and research teams. Teachers need to select appropriate article materials according to the curriculum standards and manually design or adjust the vocabulary, sentence patterns and grammar of the test questions to make them meet the understanding level of students of different grades and the examination requirements. Although this method can ensure the consistency of test question quality and teaching objectives to a certain extent, it also exposes significant problems in actual applications. First, the workload of manually designing proposition materials is huge. When teachers adapt materials for different themes and difficulties, they often need to spend a lot of time, which is difficult to continue efficiently under the condition of heavy teaching tasks. Secondly, since manual writing is prone to repetitive problems, the proposition content may lack diversity and innovation and it is difficult to stimulate students' interest in learning. In addition, the personal subjective judgment of teachers when setting questions may lead to inconsistencies in vocabulary, sentence patterns and grammar, and it is impossible to strictly ensure that all test questions meet the proposition standards for the high school entrance examination.

[0004] In recent years, some auxiliary education systems have begun to introduce intelligent tools to adjust English reading comprehension questions and materials. Auxiliary education tools usually adapt content based on keyword matching and syntactic analysis. However, such tools also have many shortcomings in practical applications. On the one hand, systems based on keyword matching can only perform surface replacements and cannot deeply understand the core themes and contextual logic of the materials, which can easily lead to the lack of coherence or deviation from the teaching objectives of the adapted content. On the other hand, the system usually lacks the ability to adaptively adjust to the junior high school English teaching syllabus and the high school entrance examination standards and cannot generate personalized test versions according to the needs of different students. In addition, the grammar and sentence rewriting modules of existing systems have limited capabilities and are difficult to handle complex clause structures and advanced grammatical expressions. There is also a lack of a complete evaluation mechanism for the quality of the generated content. Summary of the invention

[0005] One purpose of the present invention is to propose an intelligent rewriting method for reading comprehension proposition materials based on a large model, which improves teaching efficiency and ensures the quality and diversity of generated content.

[0006] According to an embodiment of the present invention, a method for intelligently rewriting reading comprehension question materials based on a large model includes the following steps:

[0007] S1. receiving an original English text input by a user, wherein the original English text includes clause structures, advanced vocabulary, and grammatical expressions beyond the cognitive level of junior high school students;

[0008] S2. Preprocess the original English text by performing word segmentation, syntactic analysis and context structure analysis, and construct a text representation model containing vocabulary information, syntactic structure information and context theme information;

[0009] S3. Using the pre-trained GPT large model, the core topic representation model and context information set of the English original text are automatically identified and extracted based on the constructed text representation model;

[0010] S4. Based on the core topic representation model and context information set, the GPT large model automatically evaluates the vocabulary, sentence structure and grammatical complexity of the English original text, and generates a text feature set including vocabulary difficulty, sentence structure complexity and grammatical complexity;

[0011] S5, GPT large model adaptively simplifies the original English text according to the junior high school English teaching syllabus and the high school entrance examination test standards, replaces advanced vocabulary with common vocabulary that meets the junior high school English teaching syllabus based on contextual information and difficulty assessment, converts complex clause structures into simple sentences or parallel sentences, and adjusts complex grammar to a grammatical structure that meets the high school entrance examination test standards based on grammatical associations;

[0012] S6. The GPT model evaluates the grammatical correctness, semantic integrity, and logical coherence of the rewritten English text generated after simplification.

[0013] S7. Based on the evaluation results and user feedback, the feedback learning module of the GPT large model dynamically adjusts the simplification rules and generation strategies. The GPT large model adaptively adjusts the simplification standards and generation strategies according to the type of English original text input by the user, the context relevance and the difficulty requirements of the proposition;

[0014] S8. Output the English text simplified by the GPT large model and meeting the standards for the high school entrance examination to the user or the education system.

[0015] Optionally, the S1 specifically includes:

[0016] S11, receiving an original English text T input by a user, the original English text including a plurality of sentences S i The sentence set S = {S1, S2, ..., S n}, each sentence S i Includes clause structure, advanced vocabulary, and grammatical expressions beyond the cognitive level of junior high school students;

[0017] S12, for each sentence S in the input English original text T i Perform grammatical analysis, separate the main clause and the subordinate clause, and construct a set of subordinate clause structures:

[0018] C i ={c i1 ,c i2 ,…,c im};

[0019] Among them, c ij Represents sentence S i The j-th clause in ;

[0020] S13. From each sentence S i Extract high-level vocabulary sets from:

[0021] V i = {v i1 ,v i2 ,…,v ik};

[0022] Among them, v ij Represents sentence S i The jth advanced vocabulary in the , advanced vocabulary refers to the vocabulary that does not match the junior high school English teaching syllabus or has a low frequency;

[0023] S14. According to each sentence S i The syntactic structure and lexical complexity of the language construct a set of grammatical expressions that exceed the cognitive level:

[0024] G i = {g i1 ,g i2 ,…,g il};

[0025] Among them, g ij Represents sentence S i The jth grammatical structure in is beyond the cognitive level of junior high school students;

[0026] S15, merging the structure, vocabulary and grammatical features of all sentences in the received English original text T to construct the English original text matrix M T :

[0027] Number of sentences Number of clauses Number of advanced vocabulary Number of grammatical structures above the level

[0028]

[0029] Among them, m i , k i , l i Respectively represent the sentence Si The number of clauses, advanced vocabulary, and grammatical structures that are beyond the students' cognitive level.

[0030] Optionally, the S2 specifically includes:

[0031] S21, for the received English original text T = {S1, S2, ..., S n} to perform word segmentation and separate each sentence S i Decompose into a sequence of lexical units:

[0032]

[0033] Among them, L i Represents sentence S i The total number of words, w ij Represents sentence S i The jth vocabulary unit in ; for each vocabulary unit w ij Perform part-of-speech tagging and vocabulary attribute extraction to obtain vocabulary feature vectors:

[0034] F ij =[pos(w ij ),lemma(w ij ),freq(w ij ),complexity(w ij )];

[0035] Among them, pos(w ij ) is the vocabulary w ij Part of speech, lemma (w ij ) is the vocabulary w ij The prototype form of freq(w ij ) is the vocabulary w ij The frequency of occurrence in the junior high school English teaching syllabus, complexity (w ij ) is the vocabulary w ij The complexity score reflects its difficulty for junior high school students;

[0036] S22. For each sentence S i Perform syntactic analysis to build a grammatical dependency tree:

[0037] D i =(N i ,E i );

[0038] in, is a set of nodes, each node n ij Corresponding vocabulary unit w ij , E i ={(n ip ,niq ,r ipq )|p,q∈[1,L i ]} is the edge set, r ipq Represents slave node n ip Points to node n iq The type of grammatical relation;

[0039] Calculate the sentence S i The syntactic complexity index C syntax (S i ):

[0040]

[0041] Among them, weight(r ipq ) is the grammatical relation r ipq The complexity weight reflects the difficulty level of the grammatical structure in the junior high school English teaching syllabus;

[0042] S23, perform context structure analysis on the original English text T, use the topic model to extract the topic distribution of the text and construct the topic distribution vector of the text:

[0043] θ T =[θ1,θ2,…,θ K ];

[0044] Among them, K is the preset number of topics, θ k represents the weight of the article T on the kth topic; at the same time, construct the topic word distribution matrix Φ = [φ kj ], where φ kj Represents the word w in the kth topic j probability;

[0045] Based on the topic distribution vector θ T And the topic word distribution matrix Φ, construct the context topic representation vector C of the chapter T ;

[0046] S24, based on vocabulary feature set F i , grammatical dependency tree D i and the contextual topic representation vector C T , construct a comprehensive text representation model R of the English original text T :

[0047] R T =(F T ,G T ,C′ T );

[0048] in, is the lexical feature set of all sentences in the text, G T =(NT ,E T ) is the global grammar graph of the text, C′ T =α1·C T +β1·Embed(T) is the weighted context topic representation vector, α1 and β1 are balance coefficients, and Embed(T) is the semantic embedding vector of passage T.

[0049] Optionally, the S3 specifically includes:

[0050] S31. Use the pre-trained GPT model to construct a comprehensive text representation model R of the original English text T Processing, automatically identifying and extracting the core topics and contextual information of the English original text;

[0051] S32, based on vocabulary feature set F T and the contextual topic representation vector C′ T , calculate the topic relevance S of the chapter topic (T):

[0052]

[0053] Among them, w j is the vocabulary feature set F T The jth vocabulary unit in θ T is the topic distribution vector of the article, φ kj is an element in the topic word distribution matrix, representing the word w in the kth topic j probability;

[0054] S33. Using grammatical dependency tree G T =(N T ,E T ), analyze the grammatical relationship between sentences and construct the sentence connectivity matrix L T =[l ij ]:

[0055]

[0056] Among them, l ij Represents sentence S i With sentence S j Connectivity

[0057] S34. Extract the main idea of ​​the article through the semantic understanding module of the GPT large model T And the key information point set K T ={k1,k2,…,k m}, where k i represents the i-th key information point;

[0058] S35. Combine the main ideas M T , key information point set K T , topic relevance S topic (T) and the sentence connectivity matrix L T , build the core topic representation model C core :

[0059] C core =(M T ,K T ,S topic (T),L T );

[0060] S36, based on the core topic representation model C core Analyze the logical relationship, narrative structure and semantic association between paragraphs in the article to generate a set of context information:

[0061]

[0062] Among them, C context Represents a set of context information, which is a quantitative representation of the logical relationship between paragraphs and the semantic association between the previous and next paragraphs. Sim(S i ,S j ) represents sentence S i and sentence S j The semantic similarity between two sentences is used to measure the semantic overlap between two sentences and to capture the potential information connection between sentences in the text. T [i,j] is an element in the sentence connectivity matrix, representing sentence S i With S j The strength of the direct grammatical relationship between topic (S i ,S j ) is a sentence S i and S j The topic relevance between them, Rel(S i ,M T ) is a sentence S i With the main idea M T , α2 and β2 are the balance coefficients between semantic similarity and topic relevance, and λ is the adjustment coefficient of topic relevance. is a logistic regression function, which is used to map the comprehensive results of sentence connectivity and topic relevance into the interval [0,1].

[0063] Optionally, the S4 specifically includes:

[0064] S41. Core topic representation model C core and context information set C context, analyze the vocabulary features of the text F T , syntactic relation G T and contextual topic information C′ T ;

[0065] S42. Use the GPT model to evaluate the vocabulary difficulty of English original texts D vocab (T), calculate each vocabulary unit w j ∈F T The lexical complexity score of is:

[0066]

[0067] Among them, complexity (w j ) represents the word w j The complexity of freq(w j ) is w j The frequency of occurrence in the junior high school English teaching syllabus, max(freq(F T )) is the maximum frequency of all words;

[0068] S43. Calculate sentence complexity D based on syntactic dependency tree syntax (S i );

[0069] S44, using context information set C context Relevance to the topic S topic (T) Calculate the grammatical complexity D of the passage grammar (T):

[0070]

[0071] Where n is the number of sentences, Rel(S i ,M T ) represents sentence S i With the main idea M T , γ and δ are the balance coefficients between sentence complexity and correlation;

[0072] S45. Generate text feature set F by integrating vocabulary, sentence structure and grammatical complexity text (T):

[0073] F text (T)=(D vocab (T),D syntax (T),D grammar (T)).

[0074] Optionally, the S5 specifically includes:

[0075] S51, according to the vocabulary feature set F Tand text feature set F text (T), identify the high-level vocabulary set V in the English original text H ={v1,v2,…,v k},in It indicates vocabulary units beyond the junior high school English syllabus;

[0076] S52, using the vocabulary replacement function Replace(v j ,V C ) Each advanced vocabulary v j ∈V H Replace with common vocabulary v′ that conforms to the junior high school English teaching syllabus j ∈V C :

[0077]

[0078] Among them, Sim(v j ,v) is the vocabulary v j and the semantic similarity of common vocabulary v, freq(v) is the frequency of v in the junior high school English teaching syllabus, V C A collection of common words;

[0079] S53, based on the syntactic dependency tree G T Identify complex clause structures in English source texts C H ={c1,c2,…,c m};

[0080] S54、Use clauses to simplify functions Simplify(c i ) Each complex clause c i ∈C H Converted into simple sentences or parallel sentences, the simplified sentence set is expressed as:

[0081] S′={s′1,s′2,…,s′ m};

[0082] Among them, s′ i For clause c i A simplified form that retains the semantic core of the original sentence but reduces syntactic complexity;

[0083] S55, using context information set C context And the sentence connectivity matrix L T , adjust the grammatical association of the simplified sentence S' to make it meet the standards of the high school entrance examination;

[0084] S56. Define a grammar adjustment function Adjust(S′) to optimize the grammatical structure of the sentence based on grammatical relevance and context information:

[0085]

[0086] Among them, S″ i For sentence s′ i The optimized form, λ is the adjustment coefficient to ensure that the optimized syntax meets the syllabus requirements;

[0087] S57, generate simplified English original text T':

[0088] T′={S″1,S″2,…,S″ n}.

[0089] The beneficial effects of the present invention are:

[0090] (1) The present invention proposes a multi-dimensional feature evaluation algorithm by combining a vocabulary feature set, a syntactic dependency tree and contextual topic information. By comprehensively calculating vocabulary complexity, sentence complexity and grammatical complexity, the present invention can automatically adjust the content of the material according to the junior high school English teaching syllabus and the high school entrance examination question standards. Compared with traditional keyword replacement, the present invention innovatively uses contextual relevance calculation functions and logistic regression functions to accurately replace complex syntax and advanced vocabulary in the material, convert complex clauses into simple sentences or parallel sentences and optimize grammatical relevance, thereby ensuring that the rewritten material not only meets the teaching standards but also maintains semantic coherence, which can effectively avoid the problems of semantic discontinuity and logical loss in the simplification process of traditional tools, and improve the accuracy and applicability of the generated content.

[0091] (2) The present invention designs a multi-layer adaptive proposition generation mechanism based on a core topic model. By analyzing the main ideas, key information points and grammatical correlation matrix between sentences in the original text, multiple versions of proposition materials can be generated according to different educational objectives. Semantic similarity and topic correlation are used as adjustment parameters to achieve full-process adaptive adjustment from text input to proposition generation. Different from the fixed template propositions commonly seen in existing tools, the present invention can flexibly generate proposition versions suitable for different difficulty levels and topics, thereby achieving diversification and innovation of proposition content.

[0092] (3) The present invention integrates dynamic feedback and real-time optimization modules so that the GPT large model can continuously optimize the generation effect according to user feedback in practical applications. Teachers or educational systems evaluate the generated materials and propositions during use. The system uses these feedbacks as training data input into the model to automatically adjust the generation strategy and algorithm parameters. Through this closed-loop optimization mechanism, the present invention can gradually improve the adaptability of the model to different student groups and realize highly personalized proposition generation, which solves the problem of lack of personalized feedback and dynamic adjustment of the model in the prior art and provides more efficient teaching support for teachers and students. BRIEF DESCRIPTION OF THE DRAWINGS

[0093] 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:

[0094] Figure 1 A flowchart of an intelligent rewriting method for reading comprehension question materials based on a large model proposed by the present invention;

[0095] Figure 2 This is a schematic diagram of the structure of an automated evaluation model for vocabulary, sentence structure and grammatical complexity in an intelligent rewriting method for reading comprehension question materials based on a large model proposed in the present invention. DETAILED DESCRIPTION

[0096] 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.

[0097] refer to Figure 1-Figure 2 , an intelligent rewriting method for reading comprehension proposition materials based on a large model, comprising the following steps:

[0098] S1. receiving an original English text input by a user, wherein the original English text includes clause structures, advanced vocabulary, and grammatical expressions beyond the cognitive level of junior high school students;

[0099] S2. Preprocess the original English text by performing word segmentation, syntactic analysis and context structure analysis, and construct a text representation model containing vocabulary information, syntactic structure information and context theme information;

[0100] S3. Using the pre-trained GPT large model, the core topic representation model and context information set of the English original text are automatically identified and extracted based on the constructed text representation model;

[0101] S4. Based on the core topic representation model and context information set, the GPT large model automatically evaluates the vocabulary, sentence structure and grammatical complexity of the English original text, and generates a text feature set including vocabulary difficulty, sentence structure complexity and grammatical complexity;

[0102] S5, GPT large model adaptively simplifies the original English text according to the junior high school English teaching syllabus and the high school entrance examination test standards, replaces advanced vocabulary with common vocabulary that meets the junior high school English teaching syllabus based on contextual information and difficulty assessment, converts complex clause structures into simple sentences or parallel sentences, and adjusts complex grammar to a grammatical structure that meets the high school entrance examination test standards based on grammatical associations;

[0103] S6. The GPT model evaluates the grammatical correctness, semantic integrity, and logical coherence of the rewritten English text generated after simplification.

[0104] S7. Based on the evaluation results and user feedback, the feedback learning module of the GPT large model dynamically adjusts the simplification rules and generation strategies. The GPT large model adaptively adjusts the simplification standards and generation strategies according to the type of English original text input by the user, the context relevance and the difficulty requirements of the proposition;

[0105] S8. Output the English text simplified by the GPT large model and meeting the standards for the high school entrance examination to the user or the education system.

[0106] In this implementation, S1 specifically includes:

[0107] S11, receiving an original English text T input by a user, the original English text including a plurality of sentences S i The sentence set S = {S1, S2, ..., S n}, each sentence S i Includes clause structure, advanced vocabulary, and grammatical expressions beyond the cognitive level of junior high school students;

[0108] S12, for each sentence S in the input English original text T i Perform grammatical analysis, separate the main clause and the subordinate clause, and construct a set of subordinate clause structures:

[0109] C i ={c i1 ,c i2 ,…,c im};

[0110] Among them, c ij Represents sentence S i The j-th clause in ;

[0111] S13. From each sentence S i Extract high-level vocabulary sets from:

[0112] V i = {v i1 ,v i2 ,…,v ik};

[0113] Among them, v ij Represents sentence S i The jth advanced vocabulary in the , advanced vocabulary refers to the vocabulary that does not match the junior high school English teaching syllabus or has a low frequency;

[0114] S14. According to each sentence S i The syntactic structure and lexical complexity of the language construct a set of grammatical expressions that exceed the cognitive level:

[0115] G i = {g i1 ,g i2 ,…,g il};

[0116] Among them, g ij Represents sentence S i The jth grammatical structure in is beyond the cognitive level of junior high school students;

[0117] S15, merging the structure, vocabulary and grammatical features of all sentences in the received English original text T to construct the English original text matrix M T :

[0118] Number of sentences Number of clauses Number of advanced vocabulary Number of grammatical structures above the level

[0119]

[0120] Among them, m i , k i , l i Respectively represent the sentence S i The number of clauses, advanced vocabulary, and grammatical structures that are beyond the students' cognitive level.

[0121] In this implementation, S2 specifically includes:

[0122] S21, for the received English original text T = {S1, S2, ..., S n} to perform word segmentation and separate each sentence S i Decompose into a sequence of lexical units:

[0123]

[0124] Among them, L i Represents sentence S i The total number of words, w ij Represents sentence S i The jth vocabulary unit in ; for each vocabulary unit w ij Perform part-of-speech tagging and vocabulary attribute extraction to obtain vocabulary feature vectors:

[0125] F ij =[pos(w ij ),lemma(w ij ),freq(w ij ),complexity(w ij )];

[0126] Among them, pos(w ij ) is the vocabulary w ijPart of speech, lemma (w ij ) is the vocabulary w ij The prototype form of freq(w ij ) is the vocabulary w ij The frequency of occurrence in the junior high school English teaching syllabus, complexity (w ij ) is the vocabulary w ij The complexity score reflects its difficulty for junior high school students;

[0127] S22. For each sentence S i Perform syntactic analysis to build a grammatical dependency tree:

[0128] D i =(N i ,E i );

[0129] in, is a set of nodes, each node n ij Corresponding vocabulary unit w ij , E i ={(n ip ,n iq ,r ipq )|p,q∈[1,L i ]} is the edge set, r ipq Represents slave node n ip Points to node n iq The type of grammatical relation;

[0130] Calculate the sentence S i The syntactic complexity index C syntax (S i ):

[0131]

[0132] Among them, weight(r ipq ) is the grammatical relation r ipq The complexity weight reflects the difficulty level of the grammatical structure in the junior high school English teaching syllabus;

[0133] S23, perform context structure analysis on the original English text T, use the topic model to extract the topic distribution of the text and construct the topic distribution vector of the text:

[0134] θ T =[θ1,θ2,…,θ K ];

[0135] Among them, K is the preset number of topics, θ k represents the weight of the article T on the kth topic; at the same time, construct the topic word distribution matrix Φ = [φ kj ], where φkj Represents the word w in the kth topic j probability;

[0136] Based on the topic distribution vector θ T And the topic word distribution matrix Φ, construct the context topic representation vector C of the chapter T ;

[0137] S24, based on vocabulary feature set F i , grammatical dependency tree D i and the contextual topic representation vector C T , construct a comprehensive text representation model R of the English original text T :

[0138] R T =(F T ,G T ,C′ T );

[0139] in, is the lexical feature set of all sentences in the text, G T =(N T ,E T ) is the global grammar graph of the text, C′ T =α1·C T +β1·Embed(T) is the weighted context topic representation vector, α1 and β1 are balance coefficients, and Embed(T) is the semantic embedding vector of passage T.

[0140] In this implementation, S3 specifically includes:

[0141] S31. Use the pre-trained GPT model to construct a comprehensive text representation model R of the original English text T Processing, automatically identifying and extracting the core topics and contextual information of the English original text;

[0142] S32, based on vocabulary feature set F T and the contextual topic representation vector C′ T , calculate the topic relevance S of the chapter topic (T):

[0143]

[0144] Among them, w j is the vocabulary feature set F T The jth vocabulary unit in θ T is the topic distribution vector of the article, φ kj is an element in the topic word distribution matrix, representing the word w in the kth topic jprobability;

[0145] S33. Using grammatical dependency tree G T =(N T ,E T ), analyze the grammatical relationship between sentences and construct the sentence connectivity matrix L T =[l ij ]:

[0146]

[0147] Among them, l ij Represents sentence S i With sentence S j Connectivity

[0148] S34. Extract the main idea of ​​the article through the semantic understanding module of the GPT large model T And the key information point set K T ={k1,k2,…,k m}, where k i represents the i-th key information point;

[0149] S35. Combine the main ideas M T , key information point set K T , topic relevance S topic (T) and the sentence connectivity matrix L T , build the core topic representation model C core :

[0150] C core =(M T ,K T ,S topic (T),L T );

[0151] S36, based on the core topic representation model C core Analyze the logical relationship, narrative structure and semantic association between paragraphs in the article to generate a set of context information:

[0152]

[0153] Among them, C context Represents a set of context information, which is a quantitative representation of the logical relationship between paragraphs and the semantic association between the previous and next paragraphs. Sim(S i ,S j ) represents sentence S i and sentence S j The semantic similarity between two sentences is used to measure the semantic overlap between two sentences and to capture the potential information connection between sentences in the text. T[i,j] is an element in the sentence connectivity matrix, representing sentence S i With S j The strength of the direct grammatical relationship between topic (S i ,S j ) is a sentence S i and S j The topic relevance between them, Rel(S i ,M T ) is a sentence S i With the main idea M T , α2 and β2 are the balance coefficients between semantic similarity and topic relevance, and λ is the adjustment coefficient of topic relevance. is a logistic regression function, which is used to map the comprehensive results of sentence connectivity and topic relevance into the interval [0,1].

[0154] In this implementation, S4 specifically includes:

[0155] S41. Core topic representation model C core and context information set C context , analyze the vocabulary features of the text F T , syntactic relation G T and contextual topic information C′ T ;

[0156] S42. Use the GPT model to evaluate the vocabulary difficulty of English original texts D vocab (T), calculate each vocabulary unit w j ∈F T The lexical complexity score of is:

[0157]

[0158] Among them, complexity (w j ) represents the word w j The complexity of freq(w j ) is w j The frequency of occurrence in the junior high school English teaching syllabus, max(freq(F T )) is the maximum frequency of all words;

[0159] S43. Calculate sentence complexity D based on syntactic dependency tree syntax (S i );

[0160] S44, using context information set C context Relevance to topic S topic (T) Calculate the grammatical complexity D of the passage grammar (T):

[0161]

[0162] Where n is the number of sentences, Rel(S i ,M T ) represents sentence S i With the main idea M T , γ and δ are the balance coefficients between sentence complexity and correlation;

[0163] S45. Generate text feature set F by integrating vocabulary, sentence structure and grammatical complexity text (T):

[0164] F text (T)=(D vocab (T),D syntax (T),D grammar (T)).

[0165] In this implementation, S5 specifically includes:

[0166] S51, according to the vocabulary feature set F T and text feature set F text (T), identify the high-level vocabulary set V in the English original text H ={v1,v2,…,v k},in It indicates vocabulary units beyond the junior high school English syllabus;

[0167] S52, using the vocabulary replacement function Replace(v j ,V C ) Each advanced vocabulary v j ∈V H Replace with common vocabulary v′ that conforms to the junior high school English teaching syllabus j ∈V C :

[0168]

[0169] Among them, Sim(v j ,v) is the vocabulary v j and the semantic similarity of common vocabulary v, freq(v) is the frequency of v in the junior high school English teaching syllabus, V C A collection of common words;

[0170] S53, based on the syntactic dependency tree G T Identify complex clause structures in English source texts C H ={c1,c2,…,c m};

[0171] S54、Use clauses to simplify functions Simplify(c i ) Each complex clause c i ∈C H Converted into simple sentences or parallel sentences, the simplified sentence set is expressed as:

[0172] S′={s′1,s′2,…,s′ m};

[0173] Among them, s′ i For clause c i A simplified form that retains the semantic core of the original sentence but reduces syntactic complexity;

[0174] S55, using context information set C context And the sentence connectivity matrix L T , adjust the grammatical association of the simplified sentence S' to make it meet the standards of the high school entrance examination;

[0175] S56. Define a grammar adjustment function Adjust(S′) to optimize the grammatical structure of the sentence based on grammatical relevance and context information:

[0176]

[0177] Among them, S″ i For sentence s′ i The optimized form, λ is the adjustment coefficient to ensure that the optimized syntax meets the syllabus requirements;

[0178] S57, generate simplified English original text T':

[0179] T′={S″1,S″2,…,S″ n}.

[0180] Embodiment 1:

[0181] In May 2024, a junior high school in a certain city was preparing for a mock high school entrance exam. The teaching team planned to select reading materials from public English resources and generate reading comprehension questions that met the requirements of the high school entrance exam. During the screening process, the English teaching and research group selected a news article from China Daily titled "Climate Change and Global Cooperation". However, the content of the article was complex and contained multiple nested clauses and advanced vocabulary such as "mitigate," "emission," and "convergence". Based on experience, the teacher judged that the language difficulty of this article was far beyond the comprehension level of junior high school students and needed to be rewritten.

[0182] In order to solve the high-intensity workload and grammatical consistency problems in the rewriting process, the teaching and research group decided to use the intelligent rewriting system based on the GPT large model of the present invention for processing. The following is the operation process of the system in a real scenario:

[0183] The teacher entered the original text fragment into the system at 15:00 on May 10, 2024, and the content is as follows:

[0184] “Efforts to mitigate climate change require convergence among nations to reduce carbon emissions and adopt sustainable practices. Failure to act in time may result in irreversible consequences.”;

[0185] The system first identified the advanced vocabulary in the article: "mitigate," "convergence," and "irreversible," and marked them as vocabulary beyond the junior high school English syllabus. The system also detected the nested clause structure in the sentence: "Efforts to mitigate climate change require convergence amongnations to reduce carbon emissions and adopt sustainable practices," which contains multiple verb phrases and conditional expressions, and the overall sentence structure is complex.

[0186] At 15:02 on May 10, 2024, the system automatically replaced advanced vocabulary with vocabulary that is easier for students to understand based on the vocabulary list in the junior high school English syllabus.

[0187] Original word: "mitigate" → replaced with "reduce";

[0188] Original word: "convergence" → replaced with "agreement";

[0189] Original word: "irreversible" → replaced with "serious";

[0190] At the same time, the system breaks down complex clause structures into multiple short sentences to improve students' comprehension ability.

[0191] Original sentence: “Efforts to mitigate climate change require convergence among nations to reduce carbon emissions and adopt sustainable practices.”

[0192] Simplified: “Countries need to work together to reduce pollution. They should also find better ways to protect the environment.”

[0193] The system generated three rewrites at 15:05 and selected the best one based on grammar, syntax, and contextual evaluation.

[0194] The system completed the automatic generation of reading comprehension questions at 15:08 on May 10, 2024, and output the following questions:

[0195] Question 1: What do countries need to do to fight climate change?

[0196] A. Use more resources;

[0197] B.Reduce pollution;

[0198] C.Ignore environmental issues;

[0199] D.Increase emissions;

[0200] Question 2: Why is it important for countries to work together?

[0201] A.To adopt new practices;

[0202] B.To agree on environmental laws;

[0203] C.To reduce serious problems;

[0204] D.To find new energy sources;

[0205] After the test questions were generated, the teaching team reviewed the test questions output by the system at 15:15 and provided feedback to the system: they hoped to further emphasize specific measures for environmental protection. The system adjusted the options and descriptions in the test questions in real time based on the feedback and generated a new version of the questions at 15:20.

[0206] Table 1 Comparison of time and efficiency between the system of the present invention and the traditional manual method in the process of processing reading comprehension proposition materials

[0207]

[0208]

[0209] Table 1 above shows that after using this system for intelligent rewriting, the teacher team used the generated test questions for mock exams. The pass rate of the system-generated test questions reached 92%, while the pass rate of the traditional test papers was 84%. In addition, the teachers were satisfied with the diversity and grammatical accuracy of the test questions generated by the system, with an average score of 9.1 (out of 10), while the average score of the traditional test questions was 7.8.

[0210] Example 1 demonstrates the efficiency and applicability of the present invention in actual teaching. Compared with traditional methods, the present invention significantly shortens the time for material processing and question generation, reduces the workload of teachers, and ensures the grammatical consistency of test questions and the diversity that meets the standards for the high school entrance examination.

[0211] The present invention proposes a multi-dimensional feature evaluation algorithm by combining a vocabulary feature set, a syntactic dependency tree and contextual topic information. By comprehensively calculating vocabulary complexity, sentence complexity and grammatical complexity, the material content can be automatically adjusted according to the junior high school English teaching syllabus and the high school entrance examination question standards. Compared with traditional keyword replacement, the present invention innovatively uses context relevance calculation functions and logistic regression functions to accurately replace complex syntax and advanced vocabulary in the material, converts complex clauses into simple sentences or parallel sentences and optimizes grammatical relevance, thereby ensuring that the rewritten material meets the teaching standards and maintains semantic coherence, which can effectively avoid the problems of semantic breaks and logical omissions in the simplification process of traditional tools, and improves the accuracy and applicability of the generated content.

[0212] The present invention designs a multi-layer adaptive proposition generation mechanism based on a core topic model. By analyzing the main ideas, key information points and grammatical correlation matrix between sentences in the original text, multiple versions of proposition materials can be generated according to different educational goals. Semantic similarity and topic correlation are used as adjustment parameters to realize adaptive adjustment of the entire process from text input to proposition generation. Different from the fixed template propositions commonly seen in existing tools, the present invention can flexibly generate proposition versions suitable for different difficulty levels and topics, realizing the diversification and innovation of proposition content.

[0213] The present invention integrates dynamic feedback and real-time optimization modules so that the GPT large model can continuously optimize the generation effect according to user feedback in practical applications. Teachers or educational systems evaluate the generated materials and propositions during use, and the system uses these feedbacks as training data to input the model to automatically adjust the generation strategy and algorithm parameters. Through this closed-loop optimization mechanism, the present invention can gradually improve the adaptability of the model to different student groups and realize highly personalized proposition generation, which solves the problem of lack of personalized feedback and dynamic adjustment of the model in the prior art and provides more efficient teaching support for teachers and students.

[0214] 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. An intelligent rewriting method for reading comprehension question materials based on a large model, characterized in that: The steps include: S1. receiving an original English text input by a user, wherein the original English text includes clause structures, advanced vocabulary, and grammatical expressions beyond the cognitive level of junior high school students; S2. Preprocess the original English text by performing word segmentation, syntactic analysis and context structure analysis, and construct a text representation model containing vocabulary information, syntactic structure information and context theme information; S3. Using the pre-trained GPT large model, the core topic representation model and context information set of the English original text are automatically identified and extracted based on the constructed text representation model; S4. Based on the core topic representation model and context information set, the GPT large model automatically evaluates the vocabulary, sentence structure and grammatical complexity of the English original text, and generates a text feature set including vocabulary difficulty, sentence structure complexity and grammatical complexity; S5, GPT large model adaptively simplifies the original English text according to the junior high school English teaching syllabus and the high school entrance examination test standards, replaces advanced vocabulary with common vocabulary that meets the junior high school English teaching syllabus based on contextual information and difficulty assessment, converts complex clause structures into simple sentences or parallel sentences, and adjusts complex grammar to a grammatical structure that meets the high school entrance examination test standards based on grammatical associations; S6. The GPT model evaluates the grammatical correctness, semantic integrity, and logical coherence of the rewritten English text generated after simplification. S7. Based on the evaluation results and user feedback, the feedback learning module of the GPT large model dynamically adjusts the simplification rules and generation strategies. The GPT large model adaptively adjusts the simplification standards and generation strategies according to the type of English original text input by the user, the context relevance and the difficulty requirements of the proposition; S8. Output the English text simplified by the GPT large model and meeting the standards for the high school entrance examination to the user or the education system.

2. According to claim 1, a method for intelligently rewriting reading comprehension question materials based on a large model is characterized in that: The S1 specifically includes: S11, receiving an original English text T input by a user, the original English text including a plurality of sentences S i The sentence set S = {S1, S2, ..., S n }, each sentence S i Includes clause structure, advanced vocabulary, and grammatical expressions beyond the cognitive level of junior high school students; S12, for each sentence S in the input English original text T i Perform grammatical analysis, separate the main clause and the subordinate clause, and construct a set of subordinate clause structures: C i ={c i1 ,c i2 ,…,c im }; Among them, c ij Represents sentence S i The j-th clause in ; S13. From each sentence S i Extract high-level vocabulary sets from: V i ={v i1 ,v i2 ,…,v ik }; Among them, v ij Represents sentence S i The jth advanced vocabulary in the , advanced vocabulary refers to the vocabulary that does not match the junior high school English teaching syllabus or has a low frequency; S14. According to each sentence S i The syntactic structure and lexical complexity of the language construct a set of grammatical expressions that exceed the cognitive level: G i ={g i1 ,g i2 ,…,g il }; Among them, g ij Represents sentence S i The jth grammatical structure in is beyond the cognitive level of junior high school students; S15, merging the structure, vocabulary and grammatical features of all sentences in the received English original text T to construct the English original text matrix M T : Number of sentences Number of clauses Number of advanced vocabulary Number of grammatical structures above the level Among them, m i , k i , l i Represents the sentence S i The number of clauses, advanced vocabulary, and grammatical structures that are beyond the students' cognitive level.

3. According to claim 1, a method for intelligently rewriting reading comprehension question materials based on a large model is characterized in that: The S2 specifically includes: S21, for the received English original text T = {S1, S2, ..., S n } to perform word segmentation and separate each sentence S i Decompose into a sequence of lexical units: Among them, L i Represents sentence S i The total number of words, w ij Represents sentence S i The jth vocabulary unit in ; for each vocabulary unit w ij Perform part-of-speech tagging and vocabulary attribute extraction to obtain vocabulary feature vectors: F ij =[pos(w ij ),lemma(w ij ),freq(w ij ),complexity(w ij )]; Among them, pos(w ij ) is the vocabulary w ij Part of speech, lemma (w ij ) is the vocabulary w ij The prototype form of freq(w ij ) is the vocabulary w ij The frequency of occurrence in the junior high school English teaching syllabus, complexity (w ij ) is the vocabulary w ij The complexity score reflects its difficulty for junior high school students; S22. For each sentence S i Perform syntactic analysis to build a grammatical dependency tree: D i =(N i ,E i ); in, is a set of nodes, each node n ij Corresponding vocabulary unit w ij , E i ={(n ip ,n iq ,r ipq )|p,q∈[1,L i ]} is the edge set, r ipq Represents slave node n ip Points to node n iq The type of grammatical relation; Calculate the sentence S i The syntactic complexity index C syntax (S i ): Among them, weight(r ipq ) is the grammatical relation r ipq The complexity weight reflects the difficulty level of the grammatical structure in the junior high school English teaching syllabus; S23, perform context structure analysis on the original English text T, use the topic model to extract the topic distribution of the text and construct the topic distribution vector of the text: i T =[θ1,θ2,…,θ K ]; Among them, K is the preset number of topics, θ k represents the weight of the article T on the kth topic; at the same time, construct the topic word distribution matrix Φ = [φ kj ], where φ kj Represents the word w in the kth topic j probability; Based on the topic distribution vector θ T And the topic word distribution matrix Φ, construct the context topic representation vector C of the chapter T ; S24, based on vocabulary feature set F i 、Grammar dependency tree D i and the contextual topic representation vector C T , construct a comprehensive text representation model R of the English original text T : R T =(F T ,G T ,C′ T ); in, is the lexical feature set of all sentences in the text, G T =(N T ,E T ) is the global grammar graph of the text, C′ T =α1·C T +β1·Embed(T) is the weighted context topic representation vector, α1 and β1 are balance coefficients, and Embed(T) is the semantic embedding vector of passage T.

4. According to claim 1, a method for intelligently rewriting reading comprehension question materials based on a large model is characterized in that: The S3 specifically includes: S31. Use the pre-trained GPT model to construct a comprehensive text representation model R of the original English text T Processing, automatically identifying and extracting the core topics and contextual information of the English original text; S32, based on vocabulary feature set F T and the contextual topic representation vector C′ T , calculate the topic relevance S of the chapter topic (T): Among them, w j is the vocabulary feature set F T The jth vocabulary unit in θ T is the topic distribution vector of the article, φ kj is an element in the topic word distribution matrix, representing the word w in the kth topic j probability; S33. Using grammatical dependency tree G T =(N T ,E T ), analyze the grammatical relationship between sentences and construct the sentence connectivity matrix L T =[l ij ]: Among them, l ij Represents sentence S i With sentence S j Connectivity S34. Extract the main idea of ​​the article through the semantic understanding module of the GPT large model T And the key information point set K T ={k1,k2,…,k m }, where k i represents the i-th key information point; S35. Combine the main ideas M T , key information point set K T , topic relevance S topic (T) and the sentence connectivity matrix L T , build the core topic representation model C core : C core =(M T ,K T ,S topic (T),L T ); S36, based on the core topic representation model C core Analyze the logical relationship, narrative structure and semantic association between paragraphs in the article to generate a set of context information: Among them, C context Represents a set of context information, which is a quantitative representation of the logical relationship between paragraphs and the semantic association between the previous and next paragraphs. Sim(S i ,S j ) represents sentence S i and sentence S j The semantic similarity between two sentences is used to measure the semantic overlap between two sentences and to capture the potential information connection between sentences in the text. T [i,j] is an element in the sentence connectivity matrix, representing sentence S i With S j The direct grammatical relationship strength between topic (S i ,S j ) is a sentence S i and S j The topic relevance between them, Rel(S i ,M T ) is a sentence S i With the main idea M T , α2 and β2 are the balance coefficients between semantic similarity and topic relevance, and λ is the adjustment coefficient of topic relevance. is a logistic regression function, which is used to map the comprehensive results of sentence connectivity and topic relevance into the interval [0,1].

5. According to claim 1, a method for intelligently rewriting reading comprehension question materials based on a large model is characterized in that: The S4 specifically includes: S41. Core topic representation model C core and context information set C context , analyze the vocabulary features of the text F T , syntactic relation G T and contextual topic information C′ T ; S42. Use the GPT model to evaluate the vocabulary difficulty of English original texts D vocab (T), calculate each vocabulary unit w j ∈F T The lexical complexity score of is: Among them, complexity (w j ) represents the word w j The complexity of freq(w j ) is w j The frequency of occurrence in the junior high school English teaching syllabus, max(freq(F T )) is the maximum frequency of all words; S43. Calculate sentence complexity D based on syntactic dependency tree syntax (S i ); S44, using context information set C context Relevance to the topic S topic (T) Calculate the grammatical complexity D of the passage grammar (T): Where n is the number of sentences, Rel(S i ,M T ) represents sentence S i With the main idea M T , γ and δ are the balance coefficients between sentence complexity and correlation; S45. Generate text feature set F by integrating vocabulary, sentence structure and grammatical complexity text (T): F text (T)=(D vocab (T),D syntax (T),D grammar (T))。 6. The intelligent rewriting method for reading comprehension question materials based on a large model according to claim 1 is characterized in that: The S5 specifically includes: S51, according to the vocabulary feature set F T and text feature set F text (T), identify the high-level vocabulary set V in the English original text H ={v1,v2,…,v k },in It indicates vocabulary units beyond the junior high school English syllabus; S52, using the vocabulary replacement function Replace(v j ,V C ) Each advanced vocabulary v j ∈V H Replace with common vocabulary v′ that conforms to the junior high school English teaching syllabus j ∈V C : Among them, Sim(v j ,v) is the vocabulary v j and the semantic similarity of common vocabulary v, freq(v) is the frequency of v in the junior high school English teaching syllabus, V C A collection of common words; S53, based on the syntactic dependency tree G T Identify complex clause structures in English source texts C H ={c1,c2,…,c m }; S54、Use clauses to simplify functions Simplify(c i ) Each complex clause c i ∈C H Converted into simple sentences or parallel sentences, the simplified sentence set is expressed as: S′={s′1,s′2,…,s′ m }; Among them, s′ i For clause c i A simplified form that retains the semantic core of the original sentence but reduces syntactic complexity; S55, using context information set C context And the sentence connectivity matrix L T , adjust the grammatical association of the simplified sentence S' to make it meet the standards of the high school entrance examination; S56. Define a grammar adjustment function Adjust(S′) to optimize the grammatical structure of the sentence based on grammatical relevance and context information: Among them, S″ i For sentence s′ i The optimized form, λ is the adjustment coefficient to ensure that the optimized syntax meets the syllabus requirements; S57, generate simplified English original text T': T′={S″1,S″2,…,S″ n }。

Images