Assisted academic writing method based on large model and teaching assistant agent

Abstract

The invention provides an auxiliary academic writing method based on a large model and a teaching assistant agent, and the method comprises the steps: inputting an academic writing requirement and a model essay generation prompt word to a pre-trained large model, and obtaining an academic writing model essay outputted by the large model; inputting the target academic paper and the fine-grained evaluation to generate cue words to a pre-trained large model to obtain fine-grained paper evaluation output by the large model; and inputting the personalized paper improvement suggestions to generate cue words to a pre-trained large model, and obtaining the personalized paper improvement suggestions output by the large model. According to the invention, writing guidance is provided for students in academic writing in a multi-mode, fine-grained and personalized manner by using a large language model, and the students are helped to improve the writing level.

Description

Technical Field

[0001] This invention relates to the field of natural language processing technology, and in particular to a large-model-based method for assisting academic writing and a teaching assistant agent. Background Technology

[0002] With the rapid development of large-scale language model technology, the text generation capabilities of language models are increasingly enhanced. Utilizing large-scale language models to assist academic writing, improve student writing quality, and serve university teaching is gaining increasing attention. However, existing large-scale language models, when assisting student writing, are firstly limited by the writing tasks themselves, often dealing with relatively superficial tasks and lacking exploration of rigorous academic writing and argumentation tasks, particularly those requiring extensive domain knowledge. Secondly, existing writing assistance methods offer coarse-grained feedback, limiting their assistance to pointing out writing problems and typically providing full-text-level evaluation feedback. They fail to provide feedback on paragraphs, sentences, or word choice, and do not proactively offer examples, reflecting their inability to cover the diverse forms of writing feedback in actual teaching scenarios. Furthermore, existing writing assistance methods are style-independent, failing to provide personalized, easily understood, and accessible language styles to support students' writing guidance based on their actual learning progress. Summary of the Invention

[0003] To address the technical problems existing in the prior art, this invention provides a large-model-based assisted academic writing method and teaching assistant agent. By utilizing a large language model, it provides students with writing guidance in academic writing in multiple ways, with fine granularity and personalization, thereby helping students improve their writing skills.

[0004] This invention provides a large-scale model-based assisted academic writing method, comprising: inputting academic writing requirements and sample essay generation prompts into a pre-trained large-scale model to obtain academic writing samples output by the large-scale model; the sample essay generation prompts are used to require the large-scale model to summarize the characteristics of the paper under the academic writing requirements through thought chain reasoning, so that the large-scale model generates the academic writing samples based on the characteristics of the paper; inputting a target academic paper and fine-grained evaluation generation prompts into the pre-trained large-scale model to obtain fine-grained paper evaluations output by the large-scale model; the target academic paper is obtained based on the academic writing requirements and the academic writing samples; the fine-grained evaluation generation prompts are used to guide the large-scale model to divide the target academic paper, so that the large-scale model generates fine-grained paper evaluations for each part of the divided paper; inputting personalized paper improvement suggestion generation prompts into the pre-trained large-scale model to obtain personalized paper improvement suggestions output by the large-scale model; the personalized paper improvement suggestion generation prompts include the target academic paper, preset personalized language style requirements, the academic writing requirements, and knowledge fragments related to the academic writing requirements.

[0005] According to the present invention, a large-scale model-based assisted academic writing method includes several example papers and corresponding human scores for each example paper. The large-scale model includes a paper summary module and an example paper generation module. The paper summary module is used to compare and learn the quality differences of several example papers through chain-of-thought reasoning, and analyze the reasons for the human scores of the example papers to summarize the characteristics of the papers under the academic writing requirements. The example paper generation module is used to generate academic writing examples based on several example papers and the characteristics of the papers, using context learning.

[0006] According to the present invention, a large-model-based method for assisting academic writing includes fine-grained evaluation prompts that include paper segmentation requirements and several example teacher evaluations. The large model includes a sliding window segmentation module and an evaluation generation module. The sliding window segmentation module is used to segment the target academic paper into several paper sections according to the paper segmentation requirements using a sliding window approach. The evaluation generation module is used to generate fine-grained paper evaluations for each paper section based on the several example teacher evaluations.

[0007] According to the present invention, a large-model-based method for assisted academic writing further includes a model fine-tuning module; the model fine-tuning module fine-tunes the large model according to a fine-tuning learning objective; the fine-tuning learning objective is:

[0008] in, For the fine-tuning learning objective, D c To fine-tune the training set size, sample papers and corresponding sample teacher evaluations are used as the fine-tuning training set. p i For the sample paper, range( LLM ( p i )) represents the number of windows to divide the space. S j For the first j One window, c ij For polishing S j Window for sample papers p i Evaluation, i The number of sample papers, j This represents the number of sliding windows.

[0009] According to a large-model-based assisted academic writing method provided by the present invention, the personalized paper improvement suggestion generation prompts also include several example papers and corresponding improvement suggestions written by human teaching assistants for the example papers; the large model includes a search keyword summary module, a search enhancement module, and a personalized paper improvement suggestion generation module; the search keyword summary module is used to summarize search keywords for each text fragment in a preset subject knowledge base; the search enhancement module is used to retrieve the most matching text fragment from the subject knowledge base as a knowledge fragment related to the academic writing requirements based on the target academic paper and the search keywords; the personalized paper improvement suggestion generation module is used to generate personalized paper improvement suggestions corresponding to the target academic paper using a context learning method based on the preset personalized language style requirements, the academic writing requirements, the knowledge fragments, the example papers, and the improvement suggestions written by human teaching assistants for the example papers; the preset personalized language style requirements are personalized language style requirements designed in the identity of a teaching assistant agent.

[0010] According to the present invention, there is a large model-based method for assisting academic writing, wherein the large model is the GLM-4 large model.

[0011] This invention also provides a teaching assistant agent for academic writing based on a large model, comprising: a first module for inputting academic writing requirements and model essay generation prompts into a pre-trained large model to obtain academic writing models output by the large model; the model essay generation prompts are used to require the large model to summarize the characteristics of the paper under the academic writing requirements through thought chain reasoning, so that the large model generates the academic writing model based on the characteristics of the paper; a second module for inputting target academic papers and fine-grained evaluation generation prompts into the pre-trained large model to obtain fine-grained paper evaluations output by the large model; the target academic paper and fine-grained evaluation prompts are used to input target academic papers and fine-grained evaluation ... pre-trained large model to obtain fine-grained paper evaluations output by the pre-trained large model to obtain target academic papers and fine-grained evaluation prompts into the pre-trained large model to obtain fine-grained paper evaluations output by the pre-trained large model to obtain target academic papers and fine-grained evaluation prompts into the pre-trained large model to obtain fine-grained paper evaluations output by the pre-trained large model to obtain target academic papers and fine-grained evaluation prompts into the pre-trained large model to obtain academic writing models; the model writing prompts are used to input target academic papers and fine-grained evaluation prompts into The academic paper is derived based on the academic writing requirements and the academic writing examples; the fine-grained evaluation generation prompts are used to guide the large model to segment the target academic paper, so that the large model generates the fine-grained paper evaluation for each segmented paper section; the third module is used to input personalized paper improvement suggestion generation prompts into the pre-trained large model to obtain personalized paper improvement suggestions output by the large model; the personalized paper improvement suggestion generation prompts include the target academic paper, preset personalized language style requirements, the academic writing requirements, and knowledge fragments related to the academic writing requirements.

[0012] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the large model-based assisted academic writing method as described above.

[0013] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the large-model-based assisted academic writing method as described above.

[0014] This invention provides a large-scale model-based method for assisted academic writing and an intelligent teaching assistant. The method includes: inputting academic writing requirements and model essays to a pre-trained large-scale model to generate prompts, resulting in model essays output by the large-scale model; inputting target academic papers and fine-grained evaluation prompts to a pre-trained large-scale model to generate fine-grained paper evaluations output by the large-scale model; and inputting personalized paper improvement suggestions to a pre-trained large-scale model to generate prompts, resulting in personalized paper improvement suggestions output by the large-scale model. This invention utilizes a large-scale language model to provide students with writing guidance in academic writing in multiple ways, with fine-grained details, and in a personalized manner, helping students improve their writing skills. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0016] Figure 1 This is a flowchart illustrating a large-model-based assisted academic writing method provided by the present invention.

[0017] Figure 2 This is a schematic diagram of the structure of a teaching assistant agent based on a large model for assisting academic writing, provided by the present invention.

[0018] Figure 3 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0020] Please refer to Figure 1 , Figure 1 This is a flowchart illustrating a large-model-based method for assisting academic writing, as provided by the present invention.

[0021] This invention provides a large-model-based method for assisting academic writing, comprising: 101: Input academic writing requirements and sample essay generation prompts into the pre-trained large model to obtain academic writing samples output by the large model; the sample essay generation prompts are used to require the large model to summarize the characteristics of the paper under the academic writing requirements through the form of thought chain reasoning, so that the large model can generate academic writing samples based on the characteristics of the paper. 102: Input the target academic paper and fine-grained evaluation generation prompts into the pre-trained large model to obtain the fine-grained paper evaluation output by the large model; the target academic paper is obtained based on academic writing requirements and academic writing examples; the fine-grained evaluation generation prompts are used to guide the large model to divide the target academic paper so that the large model can generate fine-grained paper evaluations for each part of the divided paper. 103: Input personalized paper improvement suggestions to generate prompt words into the pre-trained large model, and obtain personalized paper improvement suggestions output by the large model; the personalized paper improvement suggestions generated prompt words include the target academic paper, preset personalized language style requirements, academic writing requirements, and knowledge fragments related to academic writing requirements.

[0022] To address the challenge of leveraging large-scale language models to provide multi-faceted, fine-grained, and personalized writing guidance in professional writing courses, thereby helping students improve their writing skills, this invention generates prompts based on model essays. It produces high-quality academic writing models using these models, enabling students to complete target academic papers. Then, it generates prompts based on fine-grained evaluation, providing detailed feedback on the writing quality of the target academic paper. Finally, it generates prompts based on personalized paper improvement suggestions, offering customized improvement suggestions for the target academic paper based on the large-scale language model. This comprehensive approach supports academic writing, allows for personalized customization of writing style, and can assist students in improving their writing skills within undergraduate writing courses.

[0023] It should be noted that the prompts for generating model essays require a large-scale language model to analyze and summarize the characteristics of high-quality papers under academic writing requirements through reasoning based on thought chains. Furthermore, the large-scale language model is required to generate high-quality model essays, referencing the analyzed and summarized characteristics, while allowing the use of other example papers.

[0024] Fine-grained evaluation prompts guide a large language model to divide the target academic paper into blocks, and then local fine-grained evaluations are generated for each section of the paper after the division.

[0025] Taking into account the differences in students' knowledge levels and characteristics, personalized essay improvement suggestions generate prompts to guide the personalized essay improvement suggestions generated by the large language model to a language style that students can understand and accept, thereby further helping to improve students' writing skills.

[0026] Furthermore, personalized paper improvement suggestions can be derived by referring to fine-grained paper evaluation; fine-grained paper evaluation and personalized paper improvement suggestions can also be derived simultaneously, and this invention does not impose any particular limitations on this.

[0027] As a preferred embodiment, the sample essay generation prompts include several example papers and their corresponding human scores; the overall model includes a paper summary module and a sample essay generation module; the paper summary module is used to compare and learn the quality differences of several example papers through chain-of-thought reasoning, and analyze the reasons why the example papers received human scores in combination with the human scores of the example papers, and summarize the characteristics of papers under the requirements of academic writing; the sample essay generation module is used to generate academic writing samples based on several example papers and their characteristics, using contextual learning.

[0028] In this embodiment, academic writing requirements are first provided, along with several example papers written by students based on these requirements, and the corresponding human scores for each example paper. These are then provided to the paper summary module of the large model. The paper summary module then uses a chain-of-thought reasoning approach to compare and learn from the quality differences among the example papers and analyze the reasons for their varying human scores, thereby summarizing the characteristics of high-quality papers under these academic writing requirements. Finally, the model paper generation module, referencing the example papers and the summarized characteristics, generates high-quality academic writing models, allowing the use of excellent excerpts from the example papers.

[0029] As a preferred embodiment, the fine-grained evaluation generation prompts include paper segmentation requirements and several example teacher evaluations; the large model includes a sliding window segmentation module and an evaluation generation module; the sliding window segmentation module is used to segment the target academic paper into several paper parts according to the paper segmentation requirements; the evaluation generation module is used to generate fine-grained paper evaluations for each paper part based on the several example teacher evaluations.

[0030] Writing feedback varies from person to person. While teachers can provide writing evaluation guidelines, differences exist in the granularity and style of their feedback. In this embodiment, feedback data accumulated by teachers during teaching is anonymized and rendered harmless. Then, using prompt word engineering, a large-scale language model is employed to rewrite all the data in a stylized and unified manner, resulting in several example teacher evaluations. This helps to reduce the difficulty of academic paper evaluation tasks.

[0031] Unlike everyday writing, academic papers are often longer, making it difficult for large language models, or even those expanded from long texts, to directly generate fine-grained evaluations. In this embodiment, the sliding window segmentation module is based on a "window definition-evaluation generation" processing logic. First, the paper segmentation requirements guide the large language model to divide the target academic paper into blocks, resulting in several paper sections. Then, the evaluation generation module generates local, fine-grained evaluations for each paper section based on several example teacher evaluations. This embodiment utilizes the pre-trained knowledge of the large language model and can provide sentence- and paragraph-level evaluations and suggestions for improvement.

[0032] Large language models typically serve as helpful assistants to humans. In this embodiment, a model fine-tuning module is used to inject knowledge from the evaluation generation task in teacher assessments into the model parameters, aligning the generated results to human-level accuracy. The large model is then fine-tuned according to the fine-tuning learning objectives. The fine-tuned model is then able to generate high-quality, fine-grained evaluations for academic papers.

[0033] In a preferred embodiment, the large model also includes a model fine-tuning module; the model fine-tuning module fine-tunes the large model according to the fine-tuning learning objective; the fine-tuning learning objective is:

[0034] in, To fine-tune the learning objectives, D c To fine-tune the training set size, sample papers and corresponding sample teacher evaluations were used as the fine-tuning training set. p i For sample papers, range( LLM ( p i )) represents the number of windows to divide the space. S j For the first j One window, c ij For polishing S j Window for sample papers p i Evaluation, i The number of sample papers, j This represents the number of sliding windows.

[0035] Based on the open-source model ChatGLM3-6B from Zhipu Huazhang, this invention uses a fine-tuning training set to fine-tune a small-scale version of the ChatGLM3-6B model. This model can generate prompt words and output fine-grained paper evaluations based on fine-grained assessments. It achieves the following values ​​on automatic metrics Rouge-1, Rouge-2, Rouge-L, and BLEU-4, demonstrating excellent model training performance, as shown in Table 1.

[0036] Table 1. Fine-tuning effect table

[0037] It's important to note that Rouge is an evaluation metric used to assess the quality of automatic text summarization, machine translation, and text generation. This metric measures the quality of the generated text by calculating the lexical matching degree between the candidate summary and the reference summary. Rouge-1 measures the matching degree of 1-grams (ungrams), Rouge-2 measures the matching degree of 2-grams (bigrams), and Rouge-L measures the matching degree of the longest common subsequence. The three metrics are F1 scores. BLEU (Bilingual Evaluation Understudy) is a commonly used metric for measuring the output quality of machine translation and text generation systems. It primarily evaluates translation quality by assessing the n-gram matching between the machine translation result and the reference translation. BLEU-4 is a commonly used metric that uses 4-gram matching for measurement.

[0038] As a preferred embodiment, the personalized paper improvement suggestion generation prompts also include several example papers and corresponding improvement suggestions written by human teaching assistants for the example papers; the large model includes a search keyword summary module, a search enhancement module, and a personalized paper improvement suggestion generation module; the search keyword summary module is used to summarize search keywords for each text fragment in the preset subject knowledge base; the search enhancement module is used to retrieve the most matching text fragments from the subject knowledge base as knowledge fragments related to academic writing requirements based on the target academic paper and search keywords; the personalized paper improvement suggestion generation module is used to preset personalized language style requirements, academic writing requirements, knowledge fragments, example papers, and improvement suggestions written by human teaching assistants for the example papers, and uses context learning to generate personalized paper improvement suggestions for the corresponding target academic paper; the preset personalized language style requirements are personalized language style requirements designed by the teaching assistant agent.

[0039] Considering that generating suggestions for improving papers in professional writing tasks requires specialized background knowledge and theoretical support, and that general-purpose large-scale language models are insufficient in this regard, this embodiment first designs rules to divide the textbooks for corresponding writing courses into multiple text segments, constructing a subject knowledge base. The keyword summary module summarizes the main content of each text segment in the subject knowledge base, using it as the search keywords for that segment.

[0040] Considering the differences in students' knowledge levels and characteristics, personalized language styles that suit their understanding and acceptance are more conducive to improving their writing skills when providing paper improvement suggestions. Therefore, this embodiment divides the language style of the improvement suggestions into three different aspects, designed as a teaching assistant agent: person (first or third person), tone (lively or rigorous), and level of detail (detailed or concise). For each combination of these aspects, this embodiment designs preset personalized language style requirements for the large language model to generate paper improvement suggestions.

[0041] In practical use, academic writing requirements, target academic papers, and personalized language style requirements will be provided to the personalized paper improvement suggestion generation module. Optionally, sample papers from other students and corresponding improvement suggestions written by human teaching assistants can be provided to offer examples for the personalized paper improvement suggestion generation module.

[0042] The academic writing requirements are compared with the search keywords of each segment in the constructed knowledge base. The retrieval enhancement module calculates semantic similarity and retrieves the top knowledge base text segments with the highest semantic similarity (the most matching text segment is taken as the knowledge segment related to the academic writing requirements). Subsequently, the personalized paper improvement suggestion generation module generates personalized paper improvement suggestions for the corresponding target academic paper based on the preset personalized language style requirements, academic writing requirements, knowledge segments, example papers, and improvement suggestions written by human teaching assistants for example papers. Taking into account the writing style of human teaching assistants, it uses context learning to generate personalized paper improvement suggestions for the corresponding target academic paper.

[0043] As a preferred embodiment, the large model is the GLM-4 large model.

[0044] The GLM-4 model is a leading artificial intelligence language model developed by Zhipu AI, based on the GLM model structure proposed by the KEG Lab at Tsinghua University. It is an open bilingual (Chinese and English) bidirectional language model with strong basic capabilities, instruction following capabilities, alignment capabilities, long text processing capabilities, and text-to-graph generation capabilities.

[0045] To verify the effectiveness of this invention in improving students' writing skills, an experiment was conducted in the undergraduate writing course "Mind, Individual, and Culture" at Tsinghua University, using GLM-4 as the large-scale language model employed in this invention. First, 15 students were selected and asked to complete an initial draft of a paper based on the same requirements. Then, this invention generated a corresponding model paper to meet the requirements, and provided quality evaluation results and improvement suggestions for each student's paper. Each student was then required to revise their paper based on the model paper, the quality evaluation results, and the improvement suggestions; the revision process was iterated twice.

[0046] Subsequently, human teaching assistants were asked to grade the first draft, the first revised draft, and the second revised draft of each student's paper. The quality evaluation results of this invention and the average of the human teaching assistant scores are shown in Table 2. All scores were out of 100.

[0047] Table 2. Grading Sheet for Papers in the Course "Mind, Individual, and Culture"

[0048] This experiment shows that, with the writing assistance provided by this invention, students' writing skills are gradually improved, which confirms the conclusion that this invention can help students improve their writing skills.

[0049] To verify the effectiveness of the fine-grained essay evaluation generated by this invention in helping students, an experiment was conducted in the undergraduate writing course "Writing and Communication" at Tsinghua University, comparing it with GLM-4 and ChatGLM3-6B fine-tuning as the large-scale language models used in this invention. First, 55 students were selected and asked to complete a course essay of approximately 5000 words on the same academic topic. Then, the evaluation section of the essay was generated at the paragraph level using this invention, randomly employing one of two generation methods. The evaluation included either existing problems (fine-grained essay evaluation), directions for improvement (personalized essay improvement suggestions), or both. All students were required to provide three types of feedback for each evaluation: no help, partially helpful, and completely helpful. The results for all students are shown in Table 3, with scoring done paragraph by paragraph.

[0050] Table 3. Grading Sheet for Papers in the "Writing and Communication" Course

[0051] The results show that the fine-grained paper evaluation and personalized paper improvement suggestions provided by this invention generally offer helpful assistance to students' professional writing. Furthermore, fine-tuning ChatGLM3-6B can aid in improving academic writing, while GLM-4 provides even better support.

[0052] The teaching assistant agent for academic writing based on a large model, provided by this invention, is described below. The teaching assistant agent for academic writing based on a large model described below can be referred to in correspondence with the academic writing method for academic writing based on a large model described above.

[0053] Please refer to Figure 2 , Figure 2 This is a schematic diagram of the structure of a teaching assistant agent based on a large model for assisting academic writing, provided by the present invention.

[0054] This invention also provides a teaching assistant agent for academic writing based on a large model, comprising: a first module 1, used to input academic writing requirements and model essay generation prompts into a pre-trained large model to obtain academic writing models output by the large model; the model essay generation prompts are used to require the large model to summarize the characteristics of the paper under the academic writing requirements through thought chain reasoning, so that the large model can generate academic writing models based on the characteristics of the paper; a second module 2, used to input target academic paper and fine-grained evaluation generation prompts into a pre-trained large model to obtain fine-grained paper evaluations output by the large model; the target academic paper is obtained based on academic writing requirements and academic writing models; the fine-grained evaluation generation prompts are used to guide the large model to divide the target academic paper, so that the large model can generate fine-grained paper evaluations for each part of the divided paper; a third module 3, used to input personalized paper improvement suggestion generation prompts into a pre-trained large model to obtain personalized paper improvement suggestions output by the large model; the personalized paper improvement suggestion generation prompts include the target academic paper, preset personalized language style requirements, academic writing requirements, and knowledge fragments related to academic writing requirements.

[0055] Figure 3 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 3As shown, the electronic device may include: a processor 310, a communications interface 320, a memory 330, and a communications bus 340, wherein the processor 310, the communications interface 320, and the memory 330 communicate with each other through the communications bus 340. The processor 310 can call logic instructions in the memory 330 to execute a large-model-based assisted academic writing method. This method includes: inputting academic writing requirements and model essay generation prompts into a pre-trained large model to obtain academic writing models output by the large model; the model essay generation prompts are used to require the large model to summarize the characteristics of papers under the academic writing requirements through thought chain reasoning, so that the large model can generate academic writing models based on these characteristics; inputting a target academic paper and fine-grained evaluation generation prompts into the pre-trained large model to obtain fine-grained paper evaluations output by the large model; the target academic paper is obtained based on the academic writing requirements and academic writing models; the fine-grained evaluation generation prompts are used to guide the large model to divide the target academic paper, so that the large model can generate fine-grained paper evaluations for each divided section; inputting personalized paper improvement suggestion generation prompts into the pre-trained large model to obtain personalized paper improvement suggestions output by the large model; the personalized paper improvement suggestion generation prompts include the target academic paper, preset personalized language style requirements, academic writing requirements, and knowledge fragments related to the academic writing requirements.

[0056] Furthermore, the logical instructions in the aforementioned memory 330 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0057] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the large-model-based assisted academic writing method provided by the above methods. This method includes: inputting academic writing requirements and sample essay generation prompts into a pre-trained large model to obtain academic writing samples output by the large model; the sample essay generation prompts are used to require the large model to summarize the characteristics of the paper under the academic writing requirements through thought chain reasoning, so that the large model can generate academic writing samples based on the characteristics of the paper; inputting the target academic... The system generates prompts for the paper and fine-grained evaluations, feeding them into a pre-trained large model to obtain fine-grained paper evaluations output by the large model. The target academic paper is derived based on academic writing requirements and academic writing models. The fine-grained evaluation prompts guide the large model to segment the target academic paper, enabling the large model to generate fine-grained paper evaluations for each segmented section. The system also generates prompts for personalized paper improvement suggestions, feeding them into the pre-trained large model to obtain personalized paper improvement suggestions output by the large model. The personalized paper improvement suggestion prompts include the target academic paper, preset personalized language style requirements, academic writing requirements, and knowledge fragments related to academic writing requirements.

[0058] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the large-model-based assisted academic writing method provided by the above methods. This method includes: inputting academic writing requirements and sample essay generation prompts into a pre-trained large model to obtain academic writing samples output by the large model; the sample essay generation prompts are used to require the large model to summarize the characteristics of the paper under the academic writing requirements through thought chain reasoning, so that the large model generates academic writing samples based on the paper characteristics; inputting a target academic paper and fine-grained evaluation generation prompts into the pre-trained large model to obtain fine-grained paper evaluations output by the large model; the target academic paper is obtained based on the academic writing requirements and academic writing samples; the fine-grained evaluation generation prompts are used to guide the large model to divide the target academic paper, so that the large model generates fine-grained paper evaluations for each divided paper section; inputting personalized paper improvement suggestion generation prompts into the pre-trained large model to obtain personalized paper improvement suggestions output by the large model; the personalized paper improvement suggestion generation prompts include the target academic paper, preset personalized language style requirements, academic writing requirements, and knowledge fragments related to the academic writing requirements.

[0059] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0060] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0061] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for assisting academic writing based on a large model, characterized in that: include: Input academic writing requirements and sample essay generation prompt words into the pre-trained big model to obtain the academic writing sample essay output by the big model; The model essay generation prompt words are used to require the macro model to summarize the characteristics of the paper under the academic writing requirements in the form of thought chain reasoning, so that the macro model generates the academic writing model essay based on the characteristics of the paper; Input the target academic paper and the fine-grained evaluation generation prompt words into the pre-trained large model to obtain the fine-grained paper evaluation output by the large model; The target academic paper is obtained based on the academic writing requirements and the academic writing sample; The fine-grained evaluation generation prompt words are used to guide the large model to divide the target academic paper, so that the large model generates the fine-grained paper evaluation corresponding to each divided paper part; Inputting personalized paper improvement suggestion generation prompt words into the pre-trained large model to obtain personalized paper improvement suggestions output by the large model; The personalized paper improvement suggestion generation prompt words include the target academic paper, the preset personalized language style requirements, the academic writing requirements and the knowledge fragments related to the academic writing requirements.

2. The method for assisting academic writing based on a large model according to claim 1, characterized in that: The sample essay generation prompt includes a number of sample essays and manual scores corresponding to the sample essays; The large model includes a paper summary module and a model paper generation module; The paper summary module is used to compare and learn the quality differences of several example papers in the form of thought chain reasoning, and analyze the reasons why the example papers obtained the manual scores in combination with the manual scores of the example papers, and summarize the characteristics of the papers under the academic writing requirements; The model essay generation module is used to generate the academic writing model essay by adopting context learning method according to a number of the example essays and the features of the essays.

3. The method for assisting academic writing based on a large model according to claim 1, characterized in that: The fine-grained evaluation generation prompt words include paper division requirements and several example teacher evaluations; The large model includes a sliding window partitioning module and an evaluation generation module; The sliding window division module is used to divide the target academic paper into several paper parts by using a sliding window method according to the paper division requirements; The evaluation generation module is used to generate the fine-grained paper evaluation corresponding to each of the paper parts based on several of the example teacher evaluations.

4. The method for assisting academic writing based on a large model according to claim 3 is characterized in that: The large model also includes a model fine-tuning module; the model fine-tuning module fine-tunes the large model according to the fine-tuning learning objective; the fine-tuning learning objective is: in, For the fine-tuning learning objective, D c To fine-tune the training set size, sample papers and corresponding sample teacher evaluations are used as the fine-tuning training set. p i For the sample paper, range( LLM ( p i )) is the number of divided windows, S j For the j windows, c ij After polishing S j Sample paper under the window p i Evaluation, i is the number of sample papers, j is the number of sliding windows.

5. The method for assisting academic writing based on a large model according to claim 1, characterized in that: The personalized paper improvement suggestion generation prompt words also include several example papers and improvement suggestions written by human teaching assistants corresponding to the example papers; The large model includes a search keyword summary module, a search enhancement module, and a personalized paper improvement suggestion generation module; The search keyword summary module is used to summarize the search keywords for each text segment in the preset subject knowledge base; The search enhancement module is used to retrieve the most matching text fragment from the subject knowledge base as the knowledge fragment related to the academic writing requirement according to the target academic paper and the search keyword; The personalized paper improvement suggestion generation module is used to generate the personalized paper improvement suggestion corresponding to the target academic paper in a context learning manner according to the preset personalized language style requirements, the academic writing requirements, the knowledge fragments, the sample paper and the improvement suggestion written by the human teaching assistant of the sample paper; The preset personalized language style requirement is a personalized language style requirement designed as a teaching assistant agent.

6. The method for assisting academic writing based on a large model according to any one of claims 1 to 5, characterized in that: The large model is the GLM-4 large model.

7. A teaching assistant agent for assisting academic writing based on a large model, characterized in that: include: The first module is used to input academic writing requirements and sample essay generation prompt words into the pre-trained large model to obtain the academic writing sample essay output by the large model; The model essay generation prompt words are used to require the macro model to summarize the characteristics of the paper under the academic writing requirements in the form of thought chain reasoning, so that the macro model generates the academic writing model essay based on the characteristics of the paper; The second module is used to input the target academic paper and the fine-grained evaluation generation prompt words into the pre-trained large model to obtain the fine-grained paper evaluation output by the large model; The target academic paper is obtained based on the academic writing requirements and the academic writing sample; The fine-grained evaluation generation prompt words are used to guide the large model to divide the target academic paper, so that the large model generates the fine-grained paper evaluation corresponding to each divided paper part; The third module is used to input personalized paper improvement suggestion generation prompt words into the pre-trained large model to obtain personalized paper improvement suggestions output by the large model; the personalized paper improvement suggestion generation prompt words include the target academic paper, preset personalized language style requirements, the academic writing requirements and knowledge fragments related to the academic writing requirements.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the method for assisting academic writing based on a large model as described in any one of claims 1 to 6 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for assisting academic writing based on a large model as claimed in any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method for assisting academic writing based on a large model as claimed in any one of claims 1 to 6 is implemented.

Images