English composition correcting system based on AI large model

Abstract

The invention relates to the technical field of English homework correction, in particular to an English composition correction system based on an AI large model. Comprising the following steps: step 1, input and acquisition: inputting student compositions, directly uploading electronic editions, and uploading scanned paper boards; 2, pre-processing questions, and dividing the questions into two parts, namely, a proposal composition and a non-proposal composition according to the questions; step 3, setting a prompt project; respectively establishing judgment prompt projects according to the proposal composition and the non-proposal composition; 4, inputting the preprocessing content into an LLM large model; and step 5, the LLM model refers to the prompt project and combines data in the LLM model to judge the composition to obtain a comprehensive result. The data in the LLM model is learned and trained in advance to reach the composition judgment level for judging the examination degree. The prompt projects comprise a central thought prompt project, a question-cutting prompt project, an off-question judgment prompt project, a grammar vocabulary prompt project, a clear prompt project, a score prompt project and a guidance LLM model for judgment.

Description

Technical Field

[0001] The present invention relates to the technical field of English homework correction, and in particular to an English composition correction system based on an AI large model. Background Art

[0002] Traditional English essay grading systems mostly rely on preset rules or templates. Although these methods can handle some simple language errors, they often show limitations when faced with complex syntactic structures and diverse expressions. This rule-based system is difficult to fully understand and evaluate the semantics and logic of essays, resulting in superficial grading results and failure to deeply identify and correct students' deep-seated problems in language use.

[0003] In recent years, with the rapid development of big model technology, English essay grading systems based on big models have begun to receive widespread attention. These models are trained with massive corpora, have strong language understanding and generation capabilities, and can process complex language structures more accurately. However, the existing essay grading systems based on big models still have deficiencies in the optimization of the prompt engineering, which affects the accuracy and consistency of the correction. Insufficient optimization of the prompt engineering may result in the model not being able to fully grasp the key issues in the essay when generating correction suggestions, which in turn affects the overall user experience.

[0004] In addition, the current design of the correction result output module is rather confusing, and the feedback content lacks hierarchy and pertinence, making it difficult to provide users with meaningful improvement suggestions. This lack of refined feedback mechanism not only weakens users’ trust in the system, but also limits its application value in the field of education.

[0005] English writing is a rigid requirement for students, especially in postgraduate entrance examinations. After completing the essay, students need to know the following key information:

[0006] 1. Estimated score of the essay; 2. Specific problems in the essay; 3. Targeted improvement suggestions; 4. Personalized improvement strategies; However, the essay correction tools currently on the market have the following technical difficulties:

[0007] 1. Scoring accuracy: Traditional rule-based grading systems have difficulty accurately assessing the overall quality and score of essays.

[0008] 2. Accuracy of problem identification: Existing tools are often unable to accurately locate key problems in essays, resulting in correction results that do not meet actual needs.

[0009] 3. Relevance of suggestions: Many tools provide suggestions for improvement that are too general and lack specificity, making it difficult to help students improve effectively.

[0010] 4. Lack of personalization: The grading results are often templated and lack customized feedback based on students’ individual writing characteristics.

[0011] 5. Structured output content: The output of existing tools is often logically confusing, making it difficult to provide students with a clear path for improvement.

[0012] 6. Contextual comprehension: Especially for specific types of essays such as postgraduate entrance exams, existing tools lack an in-depth understanding of topic requirements and scoring criteria.

[0013] 7. Consistency of grading: Results from different tools or different batches of the same tool may vary greatly, affecting students’ trust.

[0014] Based on the above problems, the present invention aims to comprehensively improve the quality of AI English composition grading by innovatively applying large model technology, combined with specially optimized prompt engineering and multi-module correction structure.

[0015] By solving these technical problems, the present invention will significantly improve the practicality and effectiveness of AI English composition correction and provide students with more valuable learning support tools. Summary of the invention

[0016] The main purpose of the present invention is to provide an English composition correction system based on an AI large model so as to effectively solve the problems raised in the background technology.

[0017] To achieve the above object, the present invention adopts the following technical solution: an English composition correction system based on AI big model, comprising the following steps:

[0018] Step 1. Input and obtain: input student essays, upload the electronic version directly, or scan and upload the paper version;

[0019] Step 2: Pre-process the topic and divide it into two parts: topic-based writing and non-topic-based writing;

[0020] Step 3: Establish prompting projects; establish prompting projects for judging according to propositional essays and non-propositional essays;

[0021] Step 4: Input the above preprocessing content into the LLM large model

[0022] Step 5: The LLM model reference prompts the project, combines the data within the LLM model, and evaluates the essay to obtain a comprehensive result.

[0023] Preferably, the data in the LLM model has been trained in advance to reach the essay judgment level for judging the degree of postgraduate entrance examination.

[0024] Preferably, the prompting project for the composition in step 3 includes: a relevance prompting project, a digression judgment prompting project, a grammar and vocabulary prompting project, a clarity prompting project, and a scoring prompting project, to guide the LLM model to make judgments.

[0025] Preferably, the prompting project for the non-topic composition in step 3 includes: a central idea prompting project, a grammar and vocabulary prompting project, a clear logic prompting project, and a scoring prompting project, to guide the LLM model to make judgments.

[0026] Preferably, the prompting project described in step 3 guides the large model to correct essays in different dimensions by designing prompting content at different levels. Through refined prompting strategies, the model can independently analyze and provide feedback on grammar, vocabulary diversity, sentence structure and logic.

[0027] Preferably, the LLM model combines its own data with the requirements of the essay prompt project to judge the content of the essay. The evaluation results of the essay include: relevance results, three-dimensional results, grammar and vocabulary results, clarity results, and score results, and finally a comprehensive result is obtained.

[0028] Preferably, the LLM model combines its own data with the requirements of the non-topic composition prompt project to judge the composition content. The evaluation results of the non-topic composition include: central idea results, grammar and vocabulary results, clarity results, and score results, and finally a comprehensive result is obtained.

[0029] Preferably, the data in the LLM model can query an external knowledge base during the training and correction process, and be combined with English writing standards and style guidance related to specific fields. In the composition correction process, the LLM model can be linked with the relevant knowledge base according to the user's composition theme to provide writing background in a specific field; after obtaining the user's composition content and composition type, it will automatically query the knowledge base and perform personalized corrections based on the requirements of different composition types.

[0030] Preferably, an automatic scoring module is provided in the LLM model, a scoring mechanism based on context understanding, which can understand the context of the entire composition and then score based on the global text; the scoring mechanism is as follows:

[0031] Dbase=S×full_score+thread_hold_score where,

[0032] S is the user level automatic detection result

[0033] full_score is the full score of the current composition

[0034] thread_hold_score is based on the result of global error correction evaluation. The global error correction evaluation obtains a preliminary basic score for the student's composition and finally obtains the user's final score.

[0035] Preferably, the prompting project described in step 3 guides the large model to correct essays in different dimensions by designing prompting content at different levels. Through refined prompting strategies, the model can independently analyze and provide feedback on grammar, vocabulary diversity, sentence structure and logic.

[0036] Beneficial effects of the present invention:

[0037] 1. Comprehensiveness and accuracy: Through standardized relevance prompts, off-topic judgment prompts, grammar and vocabulary prompts, clear organization prompts, central idea prompts, and scoring prompts, the LLM model is guided to comprehensively evaluate the quality of essays.

[0038] 2. Personalization and targeting: Provide customized feedback based on students’ personalized responses, and adopt different evaluation strategies for different types of essays (propositional / non-propositional).

[0039] 3. Structured output: Clear and organized presentation of grading results, so that students can quickly understand the problems and take corrective measures.

[0040] 4. Combine the big model with a specially optimized prompting project to improve the quality of marking, with multiple modules working together to evaluate essays in all aspects.

[0041] 5. Objectivity of scoring: Based on the automatic scoring mechanism, scoring is conducted from multiple global perspectives to improve the objectivity and reliability of scoring.

[0042] 6. Solution to the large model hallucination problem: Combined with the external knowledge base, it is possible to obtain knowledge background information on different topics based on different composition topics. These background information will provide more personalized content materials for composition correction.

[0043] 7. Strong adaptability: The evaluation criteria can be adjusted according to different exam types (such as postgraduate entrance examination, TOEFL, etc.), and the system can be continuously optimized to adapt to the ever-changing writing requirements.

[0044] 8. Improve efficiency: fast and automated grading process, saving teachers and students time, real-time feedback, allowing students to make multiple revisions and study

[0045] 9. User-friendly: The intuitive interface design makes it easy for students to understand and use. It can be integrated with existing learning management systems to provide a seamless learning experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Attached Figure 1 It is a schematic diagram of the present invention. DETAILED DESCRIPTION

[0047] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0048] Attached Figure 1 As shown, an English essay correction system based on an AI big model includes the following steps:

[0049] Step 1. Input and obtain: input student essays, upload the electronic version directly, or scan and upload the paper version;

[0050] Step 2: Pre-process the topic and divide it into two parts: topic-based writing and non-topic-based writing;

[0051] Step 3: Establish prompting projects; establish prompting projects for judging according to propositional essays and non-propositional essays;

[0052] Step 4: Input the above preprocessing content into the LLM large model

[0053] Step 5: The LLM model reference prompts the project, combines the data within the LLM model, and evaluates the essay to obtain a comprehensive result.

[0054] Preferably, the data in the LLM model has been trained in advance to reach the essay judgment level for judging the degree of postgraduate entrance examination.

[0055] Preferably, the prompting project for the composition in step 3 includes: a relevance prompting project, a digression judgment prompting project, a grammar and vocabulary prompting project, a clarity prompting project, and a scoring prompting project, to guide the LLM model to make judgments.

[0056] Preferably, the prompting project for the non-topic composition in step 3 includes: a central idea prompting project, a grammar and vocabulary prompting project, a clear logic prompting project, and a scoring prompting project, to guide the LLM model to make judgments.

[0057] Preferably, the prompting project described in step 3 guides the large model to correct essays in different dimensions by designing prompting content at different levels. Through refined prompting strategies, the model can independently analyze and provide feedback on grammar, vocabulary diversity, sentence structure and logic.

[0058] Preferably, the LLM model combines its own data with the requirements of the essay prompt project to judge the content of the essay. The evaluation results of the essay include: relevance results, three-dimensional results, grammar and vocabulary results, clarity results, and score results, and finally a comprehensive result is obtained.

[0059] Preferably, the LLM model combines its own data with the requirements of the non-topic composition prompt project to judge the composition content. The evaluation results of the non-topic composition include: central idea results, grammar and vocabulary results, clarity results, and score results, and finally a comprehensive result is obtained.

[0060] Preferably, the data in the LLM model can query an external knowledge base during the training and correction process, and be combined with English writing standards and style guidance related to specific fields. In the composition correction process, the LLM model can be linked with the relevant knowledge base according to the user's composition theme to provide writing background in a specific field; after obtaining the user's composition content and composition type, it will automatically query the knowledge base and perform personalized corrections based on the requirements of different composition types.

[0061] Preferably, an automatic scoring module is provided in the LLM model, a scoring mechanism based on context understanding, which can understand the context of the entire composition and then score based on the global text; the scoring mechanism is as follows:

[0062] Dbase=S×full_score+thread_hold_score where,

[0063] S is the user level automatic detection result

[0064] full_score is the full score of the current composition

[0065] thread_hold_score is based on the result of global error correction evaluation. The global error correction evaluation obtains a preliminary basic score for the student's composition and finally obtains the user's final score.

[0066] Preferably, the prompting project described in step 3 guides the large model to correct essays in different dimensions by designing prompting content at different levels. Through refined prompting strategies, the model can independently analyze and provide feedback on grammar, vocabulary diversity, sentence structure and logic.

[0067] Beneficial effects of the present invention:

[0068] 1. Comprehensiveness and accuracy: Through standardized relevance prompts, off-topic judgment prompts, grammar and vocabulary prompts, clear organization prompts, central idea prompts, and scoring prompts, the LLM model is guided to comprehensively evaluate the quality of essays.

[0069] 2. Personalization and targeting: Provide customized feedback based on students’ personalized responses, and adopt different evaluation strategies for different types of essays (propositional / non-propositional).

[0070] 3. Structured output: Clear and organized presentation of grading results, so that students can quickly understand the problems and take corrective measures.

[0071] 4. Combine the big model with a specially optimized prompting project to improve the quality of marking, with multiple modules working together to evaluate essays in all aspects.

[0072] 5. Objectivity of scoring: Based on the automatic scoring mechanism, scoring is conducted from multiple global perspectives to improve the objectivity and reliability of scoring.

[0073] 6. Solution to the large model hallucination problem: Combined with the external knowledge base, it is possible to obtain knowledge background information on different topics based on different composition topics. These background information will provide more personalized content materials for composition correction.

[0074] 7. Strong adaptability: The evaluation criteria can be adjusted according to different exam types (such as postgraduate entrance examination, TOEFL, etc.), and the system can be continuously optimized to adapt to the ever-changing writing requirements.

[0075] 8. Improve efficiency: fast and automated grading process, saving teachers and students time, real-time feedback, allowing students to make multiple revisions and study

[0076] 9. User-friendly: The intuitive interface design makes it easy for students to understand and use. It can be integrated with existing learning management systems to provide a seamless learning experience.

[0077] Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments, or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. An English essay correction system based on AI big model, characterized by: It comprises the following steps: Step 1. Input and obtain: input student essays, upload the electronic version directly, or scan and upload the paper version; Step 2: Pre-process the topic and divide it into two parts: topic-based writing and non-topic-based writing; Step 3: Establish prompting projects; establish prompting projects for judging according to propositional essays and non-propositional essays; Step 4: Input the above preprocessed content into the LLM large model; Step 5: The LLM model reference prompts the project, combines the data within the LLM model, and evaluates the essay to obtain a comprehensive result.

2. The English essay correction system based on AI big model according to claim 1 is characterized by: The data in the LLM model has been trained in advance to reach the essay judgment level required for postgraduate entrance examinations.

3. The English essay correction system based on AI big model according to claim 1 is characterized by: The prompting project for the composition described in step 3 includes: relevance prompting project, off-topic judgment prompting project, grammar and vocabulary prompting project, clear organization prompting project, and scoring prompting project, which guide the LLM model to make judgments.

4. The English essay correction system based on AI big model according to claim 1 is characterized by: The prompting project for the non-topic composition described in step 3 includes: a central idea prompting project, a grammar and vocabulary prompting project, a clear organization prompting project, and a scoring prompting project, which guide the LLM model to make judgments.

5. The English essay correction system based on AI big model according to claim 1 is characterized by: The prompt project described in step 3 guides the large model to correct essays in different dimensions by designing prompt content at different levels. Through refined prompt strategies, the model can independently analyze and provide feedback on grammar, vocabulary diversity, sentence structure and logic.

6. The English essay correction system based on AI big model according to claim 1 is characterized by: The LLM model combines its own data with the requirements of the essay prompt project to judge the content of the essay. The evaluation results of the essay include: relevance results, three-dimensional results, grammar and vocabulary results, clarity results, and score results, and finally a comprehensive result is obtained.

7. The English essay correction system based on AI big model according to claim 1 is characterized by: The LLM model combines its own data with the requirements of the non-topic composition prompt project to judge the content of the composition. The evaluation results of the non-topic composition include: central idea results, grammar and vocabulary results, clarity results, and score results, and finally a comprehensive result is obtained.

8. The English essay correction system based on AI big model according to claim 1 is characterized by: The data in the LLM model can query external knowledge bases during the training and correction process, and be combined with English writing standards and style guidance related to specific fields. In the composition correction process, the LLM model can be linked with relevant knowledge bases according to the user's composition theme to provide writing background in specific fields; after obtaining the user's composition content and composition type, it will automatically query the knowledge base and perform personalized corrections based on the requirements of different composition types.

9. The English essay correction system based on AI big model according to claim 1 is characterized by: The LLM model is equipped with an automatic scoring module, a scoring mechanism based on contextual understanding, which can understand the context of the entire essay and then score based on the global text; the scoring mechanism is as follows: Dbase=S×full_score+thread_hold_score where, S is the user level automatic detection result full_score is the full score of the current composition thread_hold_score is based on the result of global error correction evaluation. The global error correction evaluation obtains a preliminary basic score for the student's composition and finally obtains the user's final score.

10. The English essay correction system based on AI big model according to claim 1 is characterized by: The prompt project described in step 3 guides the large model to correct essays in different dimensions by designing prompt content at different levels. Through refined prompt strategies, the model can independently analyze and provide feedback on grammar, vocabulary diversity, sentence structure and logic.

Images