English composition scoring method and device based on intelligent agent, equipment and storage medium

Abstract

The invention discloses an intelligent agent-based English composition scoring method and device, equipment and a storage medium. Comprising the following steps: inputting composition topics, composition contents and scoring standards into a scoring standard analysis agent; the scoring standard analysis agent selects a target analysis tool from a preset analysis tool group for analysis to obtain an analysis result; the analysis result comprises analysis of different scoring dimensions and corresponding suggested scores; and inputting the composition question, the composition content, the scoring standard and the analysis result into a scoring agent to obtain an English composition scoring result. Through a staged scoring mechanism, an analysis agent extracts key features to form an intermediate analysis result, and then a scoring agent performs secondary check and comprehensive judgment according to a scoring standard, so that the scoring accuracy and consistency are effectively improved. And meanwhile, the scoring standard is independently used as an external input parameter, so that the system can automatically adapt to different scoring systems, and the universality and the adaptability are enhanced.

Description

Technical Field

[0001] This application relates to the technical field of natural language processing. Specifically, it relates to a method, device, equipment and storage medium for scoring English compositions based on an agent. Background Art

[0002] Automated scoring of English compositions refers to the use of advanced computer technology and natural language processing (NLP) algorithms to achieve automated evaluation and score determination of English writing works. Its main purpose is to imitate the scoring process of human educators. By conducting a detailed analysis of various aspects of the article content - including but not limited to key dimensions such as the use of advanced vocabulary, semantic coherence, logical structure, sentence structure diversity, and topic consistency, it provides a fair and objective scoring result. Such a scoring mechanism not only significantly improves the scoring efficiency but also enables immediate feedback for learners, allowing them to quickly grasp their writing level and make targeted improvements accordingly.

[0003] Currently, scoring systems generally calculate scores comprehensively through preset quality dimension weights. For example, some systems focus on the proportional analysis of technical terms and grammar structures and meet the scoring requirements of compositions with different word count requirements and total score standards according to specific weight configurations; other systems focus on developing evaluation systems applicable to various application scenarios and achieve comprehensive evaluation using functional modules such as topic consistency detection, composition quality assessment, and coherence analysis.

[0004] Although the above technologies have made progress in improving scoring accuracy and reliability, they still face challenges in dealing with diverse scoring rules. Most existing scoring systems rely on fixed weight settings, which lack flexibility in the face of specific scoring criteria of different educational institutions or different educational levels. For example, in some cases, the richness of vocabulary may be more emphasized than the importance of logical structure, and existing systems often have difficulty quickly adapting to such changes in specific requirements. Summary of the Invention

[0005] Embodiments of this application provide a method, device, equipment and storage medium for scoring English compositions based on an agent, so as to at least solve the technical problem that the scoring system in related technologies is difficult to adapt to different scoring standards.

[0006] According to one aspect of the embodiments of this application, a method for scoring English compositions based on an agent is provided, including: Inputting the composition topic, composition content, and scoring criteria into a scoring criteria parsing agent; The scoring criteria parsing agent selects a target analysis tool from a preset group of analysis tools for parsing to obtain a parsing result; the parsing result includes the parsing of different scoring dimensions and the corresponding recommended scores; Input the composition topic, composition content, scoring criteria, and analysis results into the scoring agent to obtain the English composition scoring result.

[0007] According to another aspect of the embodiments of the present application, there is also provided an agent-based English composition scoring device, including: An input module, configured to input the composition topic, composition content, and scoring criteria into the scoring criteria analysis agent; An analysis module, configured to enable the scoring criteria analysis agent to select a target analysis tool from a preset group of analysis tools for analysis to obtain an analysis result; the analysis result includes the analysis of different scoring dimensions and the corresponding suggested scores; A scoring module, configured to input the composition topic, composition content, scoring criteria, and analysis result into the scoring agent to obtain the English composition scoring result.

[0008] According to yet another aspect of the embodiments of the present application, there is also provided an electronic device, including a memory and a processor. A computer program is stored in the memory, and the processor is configured to execute the above-mentioned agent-based English composition scoring method through the computer program.

[0009] According to yet another aspect of the embodiments of the present application, there is also provided a computer-readable storage medium. A computer program is stored in the computer-readable storage medium, wherein the computer program is configured to execute the above-mentioned agent-based English composition scoring method when running.

[0010] The technical solutions provided by the embodiments of the present application may include the following beneficial effects: Through a multi-stage scoring mechanism, the present application significantly improves the accuracy, generality, transparency, scalability, and flexibility of the English composition scoring system. Specifically, the scoring task is divided into two stages: In the first stage, the analysis agent conducts in-depth language understanding and reasoning analysis on the composition to form an intermediate analysis result; in the second stage, the scoring agent conducts a secondary verification and comprehensive judgment on the analysis result according to the scoring criteria. This mechanism is similar to the "preliminary evaluation + review" process in human expert review, effectively reducing the possibility of misjudgment by a single model and significantly improving the accuracy and consistency of scoring.

[0011] At the same time, the present application separates the scoring criteria from the model or algorithm and uses them as external input parameters, so that the system is no longer limited to a specific scoring system. Different scoring criteria documents can be provided. This design breaks the limitation that traditional scoring systems can only run according to fixed criteria and is applicable to the writing assessment needs of various examination systems, educational stages, and even different language and cultural backgrounds.

[0012] During operation, the system explicitly generates intermediate analysis results, which are re-verified and used in subsequent scoring phases. This step-by-step processing method makes the scoring path clearly visible, allowing users to trace the specific sources of each scoring metric, facilitating teachers or students to understand the scoring basis, thereby enhancing the transparency and educational feedback value of the system.

[0013] In addition, by separating the scoring criteria from the analysis process and adopting a modular design concept, this application decouples each functional component. This architecture enables new analysis tools to be independently developed and integrated according to different scoring requirements without affecting the operation logic and stability of the original analysis tools. For example, if it is necessary to add scoring dimensions for "rhetorical device recognition" or "cultural background understanding", only the corresponding analysis modules need to be developed and connected to the system, without modifying the function implementation of the existing modules. This not only improves the adaptability of the system but also greatly reduces the technical threshold and maintenance cost of function expansion, providing a solid foundation for the continuous evolution of the system in different application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The drawings described herein are used to provide a further understanding of the present application and form a part of the present application. The schematic embodiments and descriptions thereof are used to explain the present application and do not constitute an improper limitation of the present application. In the drawings: Figure 1 is a flowchart of a method for scoring English compositions based on an agent according to an embodiment of the present application; Figure 2 is a flowchart of another method for scoring English compositions according to an embodiment of the present application; Figure 3 is a flowchart of an implementation method of a scoring criterion parsing agent according to an embodiment of the present application; Figure 4 is a flowchart of an implementation of a scoring agent according to an embodiment of the present application; Figure 5 is a schematic diagram of a device for scoring English compositions based on an agent according to an embodiment of the present application; Figure 6 is a schematic diagram of the structure of an optional electronic device according to an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0015] In order to enable those skilled in the art to better understand the solution of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts shall fall within the protection scope of the present application.

[0016] It should be noted that the terms "first", "second", etc. in the description, claims and the above-mentioned drawings of this application are used to distinguish similar objects, and do not necessarily have to be used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments of this application described here can be implemented in an order other than those illustrated or described here. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0017] The following Figure 1-4 introduces in detail the agent-based English composition scoring method of the embodiments of this application.

[0018] As Figure 2 shown, a schematic diagram of an agent-based English composition scoring method is presented. It includes: a composition submission unit through which users submit English compositions to be scored. A scoring standard unit that contains the specific scoring criteria and details for guiding the scoring process. An analysis tool group that contains a variety of analysis tools used to analyze the composition in various aspects, such as grammar, vocabulary, structure, etc.

[0019] A scoring standard parsing agent that is responsible for parsing the scoring standard unit and understanding the scoring details. After the scoring standard parsing agent analyzes the composition using the analysis tool group, it generates an intermediate feature parsing result. A scoring agent that is responsible for calculating the final scoring result based on the intermediate feature parsing result, the composition submission unit, and the scoring standard unit.

[0020] As Figure 1 shown, the method mainly includes the following steps: S101 Input the composition title, composition content, and scoring criteria into the scoring standard parsing agent.

[0021] In one implementation, the composition title is the theme or question specified by the writing task, which stipulates what content the author needs to create around. The composition content is a text work written by the author based on the given title.

[0022] In one implementation, the scoring criteria include sentence structure, grammar correctness, vocabulary usage, topic relevance, content quality dimensions and corresponding scoring details in English compositions, and also include the corresponding learning stage.

[0023] The scoring criteria are scoring rules or guidelines expressed in natural language. These criteria detail the evaluation indicators of a composition in different dimensions, such as content quality, organizational structure, grammatical correctness, vocabulary usage, etc. Specific judgment scales or grades are usually given for each indicator to help evaluators objectively measure the performance of a composition according to these criteria.

[0024] It also includes the corresponding learning stage, specifically referring to a specific educational level in the education system divided according to the age, grade or cognitive development level of students. From the perspective of scoring, the learning stage is a key reference dimension when evaluating students' compositions, used to determine the specific requirements and expected levels of the scoring criteria. It reflects the degree of development of students' language abilities, thinking depth and writing skills in their learning process, thus providing a benchmark framework that matches the actual abilities of students for scoring.

[0025] The S102 scoring criteria analysis agent selects a target analysis tool from a preset group of analysis tools for analysis and obtains an analysis result; the analysis result includes the analysis of different scoring dimensions and the corresponding recommended scores.

[0026] In the embodiments of the present application, some analysis tools can be preset and stored in the group of analysis tools. The group of analysis tools includes a spelling check tool, a vocabulary richness analysis tool, a sentence structure analysis tool, a logical coherence analysis tool, a topic relevance analysis tool, an emotional color analysis tool, a grammar analysis tool, and a termination tool.

[0027] Among them, the spelling check is used to detect whether the words in the text are spelled correctly. The vocabulary richness analysis is used to evaluate the vocabulary amount and its diversity used in the text. By calculating the proportion of different words (type-token ratio), repetition rate, and identifying the use of advanced vocabulary. The sentence structure analysis is used to analyze the length, tense, complexity and structure type of sentences. The logical coherence analysis is used to evaluate the logical relationship and coherence between paragraphs and sentences. The topic relevance analysis is used to evaluate the degree of relevance of the composition content to the specified topic. The emotional color analysis is used to analyze the authenticity and sincerity of the emotions conveyed in the text. The grammar analysis is used to check the grammatical correctness in the text, including tense, voice, subject-verb agreement, etc. The specific implementation methods of the tools are not specifically limited in this application. Those skilled in the art can adaptively modify, add or reduce the tool types.

[0028] Furthermore, the scoring criteria analysis agent is mainly responsible for calculating the intermediate feature analysis results required for scoring based on the established scoring criteria using various analysis tools. Specifically, this is a method built on a large language model that, through specialized prompt words, accurately understands complex scoring rules. This agent can dynamically select and adjust the analysis tools used based on the preliminary analysis feedback to ensure that each evaluation dimension is appropriately considered. Finally, it integrates data from various tools, eliminates potential disagreements, and outputs highly accurate intermediate feature analysis results. The analysis results include the analysis of different scoring dimensions and the corresponding recommended scores. As Figure 3 shown, it specifically includes the following processes: S301 Construct task prompt words and tool prompt words, and generate the first system prompt word based on the task prompt words and tool prompt words.

[0029] S302 Serialize the composition topic, composition content, and scoring criteria to obtain the first user message.

[0030] S303 Based on the business rules corresponding to the scoring criteria, obtain a sequence of business constraint messages composed of multiple tool call messages and tool execution result messages.

[0031] In one implementation, based on the business rules corresponding to the scoring criteria, obtaining a sequence of business constraint messages composed of multiple tool call messages and tool execution result messages includes: traversing the business rules corresponding to the scoring criteria to obtain the tools to be executed; selecting the corresponding content in the composition content, composition topic, and scoring criteria as its input parameters according to the input parameter constraints of the execution tool, and constructing tool call messages; performing tool calculations based on the tools and input parameters to obtain tool execution result messages; and forming a sequence of business constraint messages with multiple tool call messages and tool execution result messages.

[0032] S304 Serialize the first system prompt word, the first user message, and the sequence of business constraint messages to obtain an input message sequence and input it into the large language model to generate the first reply message of the large language model.

[0033] S305 Identify the tool call part in the first reply message and perform deserialization to obtain the called tool and its corresponding parameters; identify the non-tool call part in the first reply message as the candidate analysis result.

[0034] S306 Execute the called tool and its corresponding parameters to obtain the tool execution result message.

[0035] S307 Add the first reply message and the tool execution result message to the input message sequence to obtain a new input message sequence, and input it into the large language model again to obtain the first reply message.

[0036] Step S308 repeats steps S305 to S307 until the tool call is terminated.

[0037] S309 Selects the candidate parsing result at the time of terminating the tool call as the parsing result for output.

[0038] S103 Inputs the composition title, composition content, scoring criteria, and parsing result into the scoring intelligent agent to obtain the English composition scoring result.

[0039] As Figure 4 shown, inputting the composition title, composition content, scoring criteria, and parsing result into the scoring intelligent agent to obtain the English composition scoring result, including: S401 Constructs a scoring task prompt as the second system prompt; the scoring task prompt includes a scoring guidance part and a format guidance part.

[0040] S402 Serializes the composition title, composition content, scoring criteria, and parsing result to obtain the second user message.

[0041] S403 Serializes the second system prompt and the second user message to obtain the input text for the large language model.

[0042] S404 Inputs the input text into the large language model to generate the second reply message of the large language model.

[0043] S405 Parses the second reply message based on the format guidance part to obtain the scoring result.

[0044] Furthermore, parsing the large language model reply message based on the format guidance part to obtain the final scoring result. If the agreed format cannot be parsed, the relevant problem information is structured into a problem message, and then the system prompt, user message input, large language model reply message, and problem message are serialized to obtain the input text for the large language model again. Repeat the text input and format parsing until a scoring result that conforms to the agreed format is generated.

[0045] To facilitate understanding of the scoring method of the embodiments of the present application. Taking the English composition scoring in junior high school stage by this method as an example, each technical detail is introduced in detail.

[0046] In junior high school, students start learning English writing. The requirements for compositions vary among different grades (seventh, eighth, and ninth grades) and different types of exams (unit tests and mid-term and final exams). For example, in the seventh grade, more emphasis is placed on the accuracy of basic grammar and word spelling. By the eighth and ninth grades, more attention is paid to the rationality of the article structure, logical coherence, and the use of complex sentence patterns. The scoring of compositions in unit tests mainly emphasizes the application of knowledge points in this unit, while mid-term and final exams focus more on evaluating whether students have achieved the overall goals of the course.

[0047] Since different schools use different textbooks, even students in the same grade may have differences in the knowledge points they learn and the content they are required to master. This leads to the diversity and complexity of composition evaluation criteria.

[0048] To meet these requirements, the automated scoring system needs to adapt to the changing scoring rules. That is, regardless of which grade, which textbook, or specific exam type, as long as the corresponding scoring rules are provided, the system can strictly follow these rules to score. The system can adapt to and accurately execute various different scoring standards, ensuring that each score is accurate and meets expectations.

[0049] In this example, it includes 1 composition submission unit and 2 scoring standard units.

[0050] Among them, the data submitted by the composition submission unit is as follows: Composition topic: Please write an English short passage titled "My Daily Life" to describe the activities and habits in your daily life.

[0051] Composition content: My daily life I'm Li Ming. I get up at six thiry. I'm walkgo to the school, we study at eight o'clock. My favorite class is English. I think it's interesting. after school study. I'm in the play groud playfootball. I'm go home. After lunch I do the homework, I'm nine o'clock go tobed. The data submitted by the two scoring standard units is as follows: Scoring criteria for the unit test scoring unit: Vocabulary Usage (30 points): Correctly use the vocabulary related to daily life learned in this unit, including but not limited to keywords such as time, activities, etc. Be able to flexibly use frequency adverbs (such as always, usually, sometimes, never, etc.) to accurately describe daily behaviors.

[0052] Sentence Structure (30 points): Be able to appropriately use a variety of sentence structures, especially the application of frequency adverbs. For example: "I usually get up at six in the morning." Pay attention to the diversity and complexity of sentences, and avoid repeating simple sentence patterns.

[0053] Content Completeness (20 points): The article should cover the main aspects of daily life, including getting up, going to school, extracurricular activities, etc., and be logically clear and information complete. The number of words should be more than 40.

[0054] Grammar Correctness (10 points): Basically have no grammar mistakes, especially the consistency of tenses and the accuracy of verb forms.

[0055] Writing Specification (10 points): The letters are written in a standard way, the words are spelled correctly, and the punctuation marks are used properly.

[0056] Learning Stage: Unit 2, Volume 1 of Grade 7 in People's Education Edition.

[0057] Scoring Criteria for the Middle School Entrance Examination Scoring Unit: Content (30 points): According to the requirements of the composition topic, the content is rich, the viewpoints are clear, all key points are covered, and there is appropriate expansion. The narration is specific, with rich details, reflecting real life situations.

[0058] Language Expression (30 points): Use words accurately, have diverse sentence patterns, be able to use relatively complex sentence structures and conjunctions to enhance the coherence of the article. Appropriately use rhetorical devices to add literary grace.

[0059] Organization Structure (20 points): The article has a reasonable structure, is well-organized, and has natural transitions. The beginning is captivating, the middle is fully discussed, and the end strongly summarizes the whole article or puts forward a prospect.

[0060] Grammar and Spelling (15 points): The grammar is correct, the spelling is error-free, and the punctuation marks are used correctly.

[0061] Creativity and Personality (5 points): Demonstrate personal style and have a certain degree of originality and innovative thinking.

[0062] Learning Stage: Junior High School.

[0063] Furthermore, the scoring criteria analysis agent conducts scoring criteria analysis.

[0064] The analysis tool library includes: Spelling Check Tool: Description: Detect whether the words in the input text are spelled correctly, and give the number of misspelled words and their corresponding error items.

[0065] Input: A composition text.

[0066] Output: The number of misspelled words and the misspelled words and their correct forms.

[0067] Lexical Richness Analysis Tool: Description: Evaluate the vocabulary used in the text and its diversity.

[0068] Input: A composition text, learning stage.

[0069] Output: Vocabulary size, vocabulary level distribution information.

[0070] Sentence Structure Analysis Tool: Description: Analyze information such as sentence length, tense, sentence type, and structure type in the composition text.

[0071] Input: A composition text, learning stage.

[0072] Output: Number of sentences, sentence pattern mastery.

[0073] Logical Coherence Analysis Tool: Description: Evaluate the logical relationship and coherence between paragraphs and sentences.

[0074] Input: A composition text.

[0075] Output: Coherence evaluation conclusion.

[0076] Thematic Relevance Analysis Tool: Description: Evaluate the degree of relevance of the composition content to the specified theme.

[0077] Input: Composition text and topic.

[0078] Output: Relevance level and its details.

[0079] Emotional Color Analysis Tool: Description: Analyze the authenticity and sincerity of the emotions conveyed in the text.

[0080] Input: Composition text.

[0081] Output: Emotional level; Grammar Analysis Tool: Description: Check the grammatical correctness in the text, including tense, voice, subject-verb agreement, etc.

[0082] Input: Composition text.

[0083] Output: Overall error rate and error entries.

[0084] Termination tool: Description: When the task is completed, select this tool to terminate the analysis process.

[0085] The implementation method of the tool in the embodiments of the present application can be implemented by using an existing analysis tool in the prior art, or a corresponding analysis tool can be set up by itself. For example, an open-source English spelling checker word-checker can be used for spelling checking. The present application does not make specific limitations on the implementation method of the tool.

[0086] Furthermore, perform scoring criterion parsing to obtain the parsing result.

[0087] First, construct a task prompt and a tool prompt, and generate a first system prompt based on the task prompt and the tool prompt.

[0088] In an exemplary scenario, the task prompt used is: You are an English composition text analysis expert who can use tools. Your task is to analyze the writing level of the provided composition. You need to reasonably use various tools according to the provided evaluation criteria, obtain the corresponding evaluation results, and conduct a summary analysis of various results to give the complete intermediate result information and conclusion required for scoring.

[0089] When using the analysis tool, please follow the following steps: 1. Select a suitable analysis tool according to the context and evaluation criteria.

[0090] 2. Provide parameters in the correct format according to the tool requirements.

[0091] 3. Observe the results and decide the next operation according to the context, evaluation criteria, and the results of the tool used.

[0092] 4. The analysis tool may change during the interaction - new tools may appear or existing tools may disappear. Please note.

[0093] Please follow the following guidelines: 1. Correctly handle errors, understand the reasons for the errors, and retry with the corrected parameters. 2. Call the tool with valid parameters according to the pattern defined in the tool documentation.

[0094] 3. If multiple tools need to be called in sequence, call one at a time and wait for the result to be returned.

[0095] 4. Please be sure to clearly explain your reasoning process and operation steps to the user.

[0096] The following is the tool information that can be used: You can find the provided tool function information in <tools>< / tools> the XML tags. For each function call, you need to return a JSON object containing the function name and parameters, and place it within the <tool_call>< / tool_call> XML tags. The tool list is as follows.

[0097] Furthermore, when obtaining the tool prompt words, first perform tool information serialization. The serialization process is to convert this structured data into text and add a special identifier. The special identifier in this instance is <tool>and< / tool> this kind of xml tag. After serializing all the tool information and then concatenating them in sequence, the tool prompt words can be obtained.

[0098] Furthermore, construct the first system prompt word. Combine the task prompt word with the tool prompt word wrapped by the special identifier to form the system prompt word. In this instance, use <tools>< / tools> this kind of xml special tag to wrap the tool prompt word and concatenate it after the task prompt word. The system prompt word for this instance can be obtained.

[0099] Furthermore, perform serialization processing on the composition topic, composition content, and grading criteria to obtain the first user message.

[0100] First, structure the composition topic, composition content, and grading criteria. The two grading criteria of this application can be respectively structured into the following two parts: Unit 2, Volume 1 of Grade 7: { "essay": { "prompt": "Please write an English short passage titled \"My Daily Life\" to describe the activities and habits in your daily life.", "content": "My daily life I'm Li Ming. I get up at six thiry. I'm walk go to the school, we study at eight o'clock. My favorite class is English. I think it's interesting. after school study. I'm in the play groud play football. I'm go home. After lunch I do the homework, I'm nine o'clock go to bed." }, "rubric": { "content": "Vocabulary usage (30 points): Correctly use the vocabulary related to daily life learned in this unit, including but not limited to keywords such as time, activities, etc. Be able to flexibly use frequency adverbs (such as always, usually, sometimes, never, etc.) to accurately describe daily behaviors.\nSentence structure (30 points): Be able to appropriately use a variety of sentence structures, especially the application of frequency adverbs, for example: “I usually get up at six in the morning.” Pay attention to the diversity and complexity of sentences and avoid repeating simple sentence patterns.\nContent integrity (20 points): The article should cover the main aspects of daily life, including getting up, going to school, after-school activities, etc., and be logically clear and information complete. The number of words should be more than 40.\nGrammar correctness (10 points): Basically no grammar mistakes, especially the consistency of tenses and the accuracy of verb forms.\nWriting norms (10 points): The letters are written in a standard way, the words are spelled correctly, and the punctuation marks are used properly.", "period": "Unit 2, Volume 1 of Grade 7, People's Education Edition" } } High school entrance examination scoring: { "essay": { "prompt": "Please write an English short passage titled \"My Daily Life\" to describe the activities and habits in your daily life.", It should be noted that there are some grammar errors in the original text, such as "I'm walk go to the school", "after school study", "I'm in the play groud play football", "I'm go home", "I'm nine o'clock go to bed", etc. The translation has been made according to the original text while trying to maintain its integrity."content": "My daily life I'm Li Ming. I get up at six thiry . I'mwalk go to the school, we study at eight o'clock . My favorite class isEnglish . I think it's interesting. after school study. I'm in the play groudplay football . I'm go home. After lunch I do the homework, I'm nine o'clockgo to bed." }, "rubric": { "content": "Content (30 points): According to the requirements of the composition topic, the content is rich, the viewpoints are clear, all key points are covered, and there is appropriate expansion. The narration is specific, with rich details, reflecting real life situations.\nLanguage expression (30 points): The words are used accurately, the sentence patterns are diverse, and relatively complex sentence structures and conjunctions can be used to enhance the coherence of the article. Appropriate rhetorical devices are used to add literary grace.\nOrganizational structure (20 points): The article structure is reasonable, well-organized, and the transitions are natural. The beginning is fascinating, the middle is fully discussed, and the end strongly summarizes the full text or presents a prospect.\nGrammar and spelling (15 points): The grammar is correct, the spelling is error-free, and the punctuation marks are used correctly.\nCreativity and personality (5 points): Show personal style, with a certain degree of originality and innovative thinking.", "period": "junior high school" } } Furthermore, perform serialization, that is, convert the above structured data into string text. Text 1: {"essay": {"prompt": "Please write an English essay titled \"My Daily Life\" to describe your activities and habits in daily life.", "content": "My daily life I'm Li Ming. I get up at six thirty . I'm walk go to the school, we study at eight o'clock . My favorite class is English . I think it's interesting. after school study. I'min the play ground play football . I'm go home. After lunch I do the homework,I'm nine o'clock go to bed."}, "rubric": {"content": "Vocabulary Use (30 points): Correctly use the vocabulary related to daily life learned in this unit, including but not limited to key vocabulary such as time and activity. Be able to flexibly use frequency adverbs (such as always, usually, sometimes, never, etc.) to accurately describe daily behaviors.\nSentence Structure (30 points): Be able to use a variety of sentence structures appropriately, especially those involving frequency adverbs, for example: "I usually get upat six in the morning." Pay attention to the diversity and complexity of sentences, and avoid repeating simple sentences. \nContent completeness (20 points): The article should cover the main aspects of daily life, including getting up, going to school, extracurricular activities, etc., and the logic should be clear and the information should be complete. The number of words should be more than 40 words. \nGrammatical correctness (10 points): There are basically no grammatical errors, especially the consistency of tenses and the accuracy of verb forms. \nWriting standard (10 points): The letters are written in a standard way, the words are spelled correctly, and the punctuation is used appropriately. ", "period": "People's Education Grade 7 Volume 1 Unit 2"}}.

[0101] Text 2: {"essay": {"prompt": "Please write an English short essay titled \"My Daily Life\" to describe the activities and habits in your daily life.", "content": "My daily life I'm Li Ming. I get up atsix thiry . I'm walk go to the school, we study at eight o'clock . Myfavorite class is English . I think it's interesting. after school study. I'min the play groud play football . I'm go home. After lunch I do the homework,I'm nine o'clock go to bed."}, "rubric": {"content": "Content (30 points): According to the requirements of the composition topic, the content is rich, the viewpoints are clear, all key points are covered, and there is appropriate expansion. The narration is specific, with rich details, reflecting real life situations.\nLanguage expression (30 points): The words are used accurately, the sentence patterns are diverse, and relatively complex sentence structures and conjunctions can be used to enhance the coherence of the article. Appropriate rhetorical devices are used to add literary grace.\nOrganizational structure (20 points): The article structure is reasonable, well-organized, and the transition is natural. The beginning is fascinating, the middle is fully discussed, and the end strongly summarizes the full text or puts forward a prospect.\nGrammar and spelling (15 points): The grammar is correct, the spelling is correct, and the punctuation marks are used correctly.\nCreativity and personality (5 points): Show personal style, with a certain degree of originality and innovative thinking.", "period": "junior high school"}}。

[0102] Obtain the user message based on the serialized result.

[0103] Furthermore, based on the business rules corresponding to the scoring criteria, obtain a sequence of business constraint messages composed of multiple tool call messages and tool execution result messages.

[0104] First, traverse the business rules corresponding to the scoring criteria to obtain the tools to be executed; according to the input parameter constraints of the execution tools, select the corresponding content in the composition content, composition title, and scoring criteria as its input parameters to construct a tool call message; according to the tool and input parameters, execute the tool calculation to obtain a tool execution result message; form a sequence of business constraint messages with multiple tool call messages and tool execution result messages.

[0105] Further, serialize the first system prompt, the first user message, and the business constraint message sequence to obtain an input message sequence and input it into the large language model to generate the first response message of the large language model.

[0106] In this example, the open-source large language model Qwen2.5-32B-Instruct is used as the large language model.

[0107] Further, sequentially concatenate the system prompt, the user message, and the business constraint message sequence into an input message sequence. The input message sequence is then processed according to the chat template of the Qwen2.5-32B-Instruct model to form the input of the large language model, which is input into the large language model to generate the response message of the large language model.

[0108] Further, parse the response message to obtain the parsing result.

[0109] Specifically, identify the tool call part in the response of the large language model and deserialize it to obtain the tool and its parameters. At the same time, regard the non-call part as the candidate parsing result, and execute the tool call to obtain the execution result message.

[0110] Add the response message of the large language model and the tool execution result message to the input message sequence to form a new input sequence, and then input it into the large language model again, repeating the above process.

[0111] When the termination tool is called, select the candidate parsing result at this time as the final parsing result for output.

[0112] In one implementation, it includes: e. Input the input text into the large language model to generate the response message of the large language model; f. Parse the response message of the large language model. It includes: 1. Identify the tool call part in the generation result of the large language model by rules and deserialize it to obtain the tool call and its corresponding parameters; 2. Identify the non-tool call part in the generation result of the large language model by rules as the candidate parsing result; g. Execute the tool call and its corresponding parameters to obtain the corresponding tool result, and process the tool result into the tool execution result message; h. Add the response message of the large language model in f and the tool execution result message in g to the message sequence in d, then serialize it to obtain the input text of the large language model, and then execute steps e - h until the tool call is terminated; i. Select the candidate parsing result at the time of terminating the tool call as the parsing result of the intermediate feature for output.

[0113] In this example, the above two inputs will further yield the following output results: Analysis results of evaluation criteria: In terms of vocabulary usage, the composition to be scored demonstrated a reasonable level, using 39 words. The analysis shows that its vocabulary usage reached an excellent standard. Given that the full score for this part is 30 points, it is recommended to give a score of 25 to 28 points, which is more appropriate.

[0114] In terms of sentence patterns, the composition mainly adopted two simple sentence patterns, without the application of complex clauses or tenses, lacking diverse sentence expressions. Considering that the author is in the learning stage of the second unit in the first volume of the seventh grade, the requirements for this part are relatively loose, but the monotony of the sentence patterns in the text is still insufficient. It is recommended to give a score of about 15 points.

[0115] Regarding the integrity of the content, the composition used a total of 57 words, which could completely cover the description of a day's itinerary. Moreover, through the evaluation of the analysis tool, it was confirmed that the content of the article highly matched the requirements of the topic. Considering the current learning stage of the student, it is recommended to give a score of 15 to 20 points.

[0116] In terms of grammar correctness, there were 5 grammar mistakes in the composition, mainly some basic mistakes. Given that the student is in the stage of learning these basic grammar rules and has not fully mastered the relevant knowledge, it is recommended to give a score of 3 to 5 points.

[0117] Finally, in terms of writing norms, since there is a lack of appropriate tools for precise analysis, it is recommended to give a full score of 10 points.

[0118] Analysis results of evaluation criteria: In terms of content (30 points): The composition to be scored basically met the requirements of the topic in terms of theme, but the content was relatively thin, the vocabulary was far lower than the level expected by the scoring criteria, lacking specific details and substantial expressions, and the overall performance was poor. It is recommended that the score range be 10 - 14 points.

[0119] Language expression (30 points): A large number of basic vocabulary was used in the article, the sentence patterns were single, mainly relying on two simple sentence patterns, lacking the application of language structures such as clauses and tense changes, and the overall language ability was far from meeting the basic requirements of the junior high school stage, belonging to a very low level. It is recommended that the score range be 3 - 5 points.

[0120] Organizational structure (20 points): The composition basically followed the chronological order in narration, but lacked necessary conjunctions and logical transitions, the connection between paragraphs was not tight, and the overall structure was relatively loose, belonging to the lower-middle level. It is recommended that the score range be 5 - 7 points.

[0121] Grammar and Spelling (15 points): There are 5 grammar mistakes in the text, mainly including basic errors, and also two spelling mistakes. These mistakes should be avoided as much as possible in junior high school English writing, which reflects the author's weak language foundation and extremely poor mastery level. The recommended score range is 1 - 3 points.

[0122] Creativity and Personal Expression (5 points): The content of the composition is mediocre, lacking novelty and personal style, belonging to a "running account" type of expression. As an additional point item, this composition fails to show any advantages, and the recommended score is 0 points.

[0123] Furthermore, the scoring agent conducts the scoring.

[0124] First, construct the scoring task prompt as the second system prompt; the scoring task prompt includes a scoring guidance part and a format guidance part.

[0125] The scoring guidance part includes: You are an intelligent scoring assistant, responsible for scoring students' compositions according to the preset scoring criteria and analysis results. Please strictly follow the following rules and steps: 1. Combine the student's current learning progress, understand and accurately grasp the evaluation scale of each indicator in the preset scoring criteria.

[0126] 2. Clearly master the key judgment basis in the scoring criteria to ensure that the scoring logic is consistent and well-founded.

[0127] 3. Combine the various analysis conclusions and preliminary scoring suggestions provided in the "intermediate feature integration result", and verify them against the scoring criteria to determine whether to adopt or adjust the suggestion.

[0128] 4. Finally, output the total score of the composition and clearly list the scoring basis to ensure that the scoring process is rigorous, transparent, and interpretable.

[0129] The format guidance part includes: Your output format needs to be in json format and use <output>< / output> to wrap the output result. The output json needs to contain two keys, namely score: used to represent the total score, and its value data type is an integer. And detail: the scoring basis, and its value data type is a str.

[0130] The above two are concatenated in order to obtain the second system prompt.

[0131] Furthermore, serialize the composition title, composition content, scoring criteria, and analysis results to obtain the second user message.

[0132] Further, serialize the second system prompt and the second user message to obtain the input text for the large language model.

[0133] Further, input the input text into the large language model to generate the second response message of the large language model.

[0134] Finally, parse the second response message based on the format guidance section to obtain the scoring result.

[0135] Specifically, parse the response message according to the format of the system prompt. It includes: a. Extract through regular expressions <output>< / output> The wrapped content; b. Then deserialize the extracted content to obtain a JSON-formatted structure; c. Check whether "score" and "detail" are in the structure; d. Check whether the value of "score" is of integer type and whether the value of "detail" is of string type; If any of the above steps fails, it means that the agreed format cannot be parsed.

[0136] If there is a format parsing error, a problem message will be constructed. The problem message consists of two parts, the role and the content. Here, the role of the problem message body is "tool", and the content is "[Format parsing failed]" concatenated with the reason for failure. For example, format parsing failed: the fields \"score\" and \"detail\" do not exist in the result.

[0137] Then serialize the system prompt, user message, large language model response message, and problem message to obtain the input text for the large language model again, and repeat the steps of message input and parsing until a text that conforms to the agreed format is generated.

[0138] In the embodiment of the present application, the first scoring result obtained is: { "score": 68, "detail": "The score for vocabulary usage is 25 points. Because daily life-related vocabulary is reasonably used in the composition, but there are a few errors; the score for sentence structure is 15 points. Because the sentence patterns are relatively single and lack complexity and variation; the score for content integrity is 18 points. Because it covers multiple aspects of daily life and the information is relatively complete; the score for grammar correctness is 5 points. Because there are several basic grammar errors but they do not seriously affect understanding; the score for writing norms is 10 points. Because there are no obvious spelling or punctuation problems. The total score is 25 + 15 + 18 + 5 + 10 = 68 points." } The second scoring result obtained is: { "score": 26, "detail": "In terms of content (12 points): The composition generally revolves around the topic and describes the main activities of daily life, but the content is relatively thin, lacking details and expansion. In terms of language expression (4 points): A large number of basic vocabulary is used, the sentence patterns are single, lacking complex sentence structures and tense changes. In terms of organizational structure (6 points): The narration generally follows the chronological order, but lacks connecting words and the logical transition is not natural. In terms of grammar and spelling (2 points): There are multiple grammar errors and spelling mistakes, and the basic language skills are weak. In terms of creativity and personality (0 points): The content is plain and lacks novelty, presenting a running account style of expression." } It can be seen that for the solution of this application, for the same composition, by inputting different evaluation criteria, the scoring results under the corresponding criteria can be obtained, proving that this method can adapt to different evaluation criteria.

[0139] The English composition scoring method of this application has achieved remarkable beneficial effects in many aspects: Improve scoring accuracy: Adopt a two-stage scoring mechanism. In the first stage, the parsing agent extracts the parsing results, and in the second stage, the scoring agent conducts a secondary check and comprehensive judgment. This "preliminary evaluation + review" process effectively reduces the possibility of misjudgment by a single model and significantly improves the accuracy and consistency of scoring.

[0140] Enhance generality and adaptability: Make the scoring criteria an external input parameter, enabling the system to automatically adapt to different scoring systems. Regardless of changes in the examination system, educational stage, or language and cultural background, as long as the corresponding scoring criteria document is provided, the system can quickly adjust the scoring dimensions and weights, breaking the limitations of traditional systems.

[0141] Improve transparency and interpretability: The system explicitly generates intermediate analysis results during operation, such as grammar correctness, paragraph cohesion, etc., and re-checks them during subsequent scoring. This step-by-step processing method makes the scoring path clearly visible, facilitating teachers and students to understand the scoring basis and enhancing the transparency and educational feedback value of the system.

[0142] Improve scalability and flexibility: Adopt a modular design, decouple the scoring criteria from the analysis process, enabling new analysis tools to be independently developed and integrated without affecting the stability of the original modules. This architecture not only improves the adaptability of the system but also reduces the technical threshold and maintenance cost of function expansion, providing a solid foundation for the continuous evolution of the system.

[0143] According to another aspect of the embodiments of this application, there is also provided an agent-based English composition scoring device for implementing the above-mentioned agent-based English composition scoring method. As Figure 5As shown in the figure, the device includes: An input module 501, configured to input the composition topic, composition content, and grading criteria into the grading criteria analysis agent; An analysis module 502, configured to enable the grading criteria analysis agent to select a target analysis tool from a preset group of analysis tools for analysis to obtain an analysis result; the analysis result includes the analysis of different grading dimensions and the corresponding recommended scores; A grading module 503, configured to input the composition topic, composition content, grading criteria, and analysis result into the grading agent to obtain an English composition grading result.

[0144] It should be noted that when the above-described agent-based English composition grading device executes the agent-based English composition grading method, only the above-mentioned division of each functional module is used as an example for illustration. In practical applications, the above functions can be allocated to different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the agent-based English composition grading device provided in the above embodiment and the agent-based English composition grading method embodiment belong to the same concept. The implementation process is detailed in the method embodiment and will not be repeated here.

[0145] According to another aspect of the embodiments of the present application, an electronic device corresponding to the agent-based English composition grading method provided in the foregoing embodiments is further provided to execute the above-mentioned agent-based English composition grading method.

[0146] Please refer to Figure 6 , which shows a schematic diagram of an electronic device provided in some embodiments of the present application. As Figure 6 shown, the electronic device includes: a processor 600, a memory 601, a bus 602, and a communication interface 603. The processor 600, the communication interface 603, and the memory 601 are connected through the bus 602; a computer program that can run on the processor 600 is stored in the memory 601, and when the processor 600 runs the computer program, it executes the agent-based English composition grading method provided in any one of the foregoing embodiments of the present application.

[0147] Among them, the memory 601 may include a high-speed random access memory (RAM: Random Access Memory), and may also include a non-volatile memory, such as at least one disk memory. Through at least one communication interface 603 (which can be wired or wireless), a communication connection between the system network element and at least one other network element is realized, and the Internet, wide area network, local area network, metropolitan area network, etc. can be used.

[0148] The bus 602 can be an ISA bus, a PCI bus, an EISA bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. Among them, the memory 601 is used to store programs. After receiving an execution instruction, the processor 600 executes the program. Any implementation manner of the agent-based English composition scoring method disclosed in any implementation manner of the embodiments of the present application can be applied to the processor 600 or implemented by the processor 600.

[0149] The processor 600 may be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method can be completed by the integrated logic circuit in the hardware of the processor 600 or by instructions in software form. The above-mentioned processor 600 can be a general-purpose processor, including a central processing unit (CPU for short), a network processor (NP for short), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. It can implement or execute the various methods, steps, and logic block diagrams disclosed in the embodiments of the present application. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in combination with the embodiments of the present application can be directly embodied as being executed and completed by a hardware decoding processor, or by a combination of hardware and software modules in the decoding processor. The software module can be located in a mature storage medium in the art such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. This storage medium is located in the memory 601, and the processor 600 reads the information in the memory 601 and combines its hardware to complete the steps of the above method.

[0150] The electronic device provided in the embodiments of the present application and the agent-based English composition scoring method provided in the embodiments of the present application are based on the same inventive concept and have the same beneficial effects as the methods adopted, run, or implemented by them.

[0151] According to another aspect of the embodiments of the present application, there is also provided a computer-readable storage medium corresponding to the agent-based English composition scoring method provided in the foregoing embodiments, on which a computer program (i.e., a program product) is stored. When the computer program is run by a processor, it will execute the agent-based English composition scoring method provided in any of the foregoing embodiments.

[0152] It should be noted that examples of computer-readable storage media may also include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory, or other optical and magnetic storage media, which will not be elaborated here one by one.

[0153] The computer-readable storage medium provided by the above embodiments of the present application and the agent-based English composition scoring method provided by the embodiments of the present application are based on the same inventive concept and have the same beneficial effects as the methods adopted, run, or implemented by the application programs stored therein.

[0154] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0155] The above embodiments only represent several implementation manners of the present invention, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent for the present invention. It should be noted that for those of ordinary skill in the art, without departing from the inventive concept of the present invention, several modifications and improvements can still be made, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent for the present invention should be subject to the appended claims.

Claims

1. An agent-based English composition scoring method, characterized in that, including: inputting the composition topic, composition content, and scoring criteria into the scoring criteria analysis agent; the scoring criteria analysis agent selects a target analysis tool from a preset group of analysis tools for analysis to obtain an analysis result; the analysis result includes the analysis of different scoring dimensions and the corresponding suggested scores; inputting the composition topic, composition content, scoring criteria, and analysis result into the scoring agent to obtain the English composition scoring result.

2. The agent-based English composition scoring method according to claim 1, wherein The scoring criteria analysis agent selects a target analysis tool from a preset group of analysis tools for analysis to obtain an analysis result, including: constructing a task prompt and a tool prompt, and generating a first system prompt based on the task prompt and the tool prompt; serializing the composition topic, composition content, and scoring criteria to obtain a first user message; obtaining a sequence of business constraint messages composed of multiple tool call messages and tool execution result messages based on the business rules corresponding to the scoring criteria; serializing the first system prompt, the first user message, and the sequence of business constraint messages to obtain an input message sequence and inputting it into the large language model to generate a first reply message of the large language model; analyzing the first reply message to obtain an analysis result.

3. The agent-based English composition scoring method according to claim 2, wherein Analyzing the first reply message to obtain an analysis result, including: a. Identifying the tool call part in the first reply message and deserializing it to obtain the called tool and its corresponding parameters; identifying the non-tool call part in the first reply message as the candidate analysis result; b. Executing the called tool and its corresponding parameters to obtain a tool execution result message; c. Adding the first reply message in step a and the tool execution result message in step b to the input message sequence to obtain a new input message sequence, and inputting it into the large language model again to obtain a first reply message; d. Repeating steps a to c until the tool call is terminated; e. Selecting the candidate analysis result at the time of terminating the tool call as the analysis result for output.

4. The agent-based English composition scoring method according to claim 2, wherein Obtaining a sequence of business constraint messages composed of multiple tool call messages and tool execution result messages based on the business rules corresponding to the scoring criteria, including: traversing the business rules corresponding to the scoring criteria to obtain the tools to be executed; selecting the corresponding content in the composition content, composition topic, and scoring criteria as its input parameters according to the input parameter constraints of the executed tool to construct a tool call message; calculating by executing the tool according to the tool and the input parameters to obtain a tool execution result message; forming a sequence of business constraint messages with multiple tool call messages and tool execution result messages.

5. The agent-based English composition scoring method according to claim 1, wherein, Inputting the composition topic, composition content, scoring criteria, and analysis result into the scoring agent to obtain the English composition scoring result, including: constructing a scoring task prompt as the second system prompt; the scoring task prompt includes a scoring guidance part and a format guidance part; serializing the composition topic, composition content, scoring criteria, and analysis result to obtain a second user message; serializing the second system prompt and the second user message to obtain the input text of the large language model; inputting the input text into the large language model to generate a second reply message of the large language model; Parse the second reply message based on the format guidance part to obtain the scoring result.

6. The agent-based English composition scoring method according to claim 1, wherein The scoring criteria include sentence structure, grammar correctness, vocabulary usage, topic relevance, content quality dimensions and corresponding scoring rules in English compositions, and also include the corresponding learning stage.

7. The agent-based English composition scoring method according to claim 1, wherein The analysis tool group includes a spelling check tool, a vocabulary richness analysis tool, a sentence structure analysis tool, a logical coherence analysis tool, a topic relevance analysis tool, an emotional color analysis tool, a grammar analysis tool, and a termination tool.

8. An agent-based English composition scoring device, characterized in that, including: An input module for inputting the composition topic, the composition content, and the scoring criteria into the scoring criteria parsing agent; A parsing module for the scoring criteria parsing agent to select a target analysis tool from a preset analysis tool group for parsing to obtain a parsing result; the parsing result includes the parsing of different scoring dimensions and the corresponding recommended scores; A scoring module for inputting the composition topic, the composition content, the scoring criteria, and the parsing result into the scoring agent to obtain the English composition scoring result.

9. An electronic device, characterized in that, It includes a processor and a memory storing program instructions, and the processor is configured to execute the agent-based English composition scoring method according to any one of claims 1 to 7 when executing the program instructions.

10. A computer-readable medium, characterized in that, Computer-readable instructions are stored thereon, and the computer-readable instructions are executed by a processor to implement an agent-based English composition scoring method according to any one of claims 1 to 7.

Images