An essay intelligent evaluation method and device

Abstract

The application belongs to the technical field of data processing, and particularly relates to a composition intelligent evaluation method and device. The method comprises the following steps: obtaining the composition content displayed by a user in a scanning area through a scanning device; respectively evaluating each sentence of the composition in terms of vocabulary, semantics, rhythm and novelty, and calculating the score of each sentence by weighting; when the score of a sentence exceeds a set value, obtaining an evaluation sentence obtained by a large model in terms of vocabulary, semantics, rhythm and novelty; respectively evaluating the composition theme, structure, language, thinking and specification based on the large model, and calculating the composition score by weighting; correcting the composition score according to a correction formula; projecting the corrected composition score to the scanning area, and projecting the evaluation sentence to the corresponding sentence side of the scanning area. The application realizes fine evaluation of the composition, and improves the objectivity and teaching pertinence of the evaluation.

Description

Technical Field

[0001] This application belongs to the field of data processing technology, and specifically relates to an intelligent essay evaluation method and device. Background Technology

[0002] In the field of Chinese language education, essay assessment is a crucial part of the teaching process. Traditional methods rely primarily on manual teacher correction, which suffers from low efficiency, strong subjectivity, and long feedback cycles, failing to meet students' needs for frequent writing practice. With the development of natural language processing technology, rule-based or statistical model-based automated essay scoring systems are gradually being applied in teaching scenarios, but existing solutions still have significant shortcomings. Most systems only provide overall scores based on superficial features such as grammatical correctness and vocabulary richness, lacking a comprehensive evaluation of sentences in terms of vocabulary usage, semantic depth, rhythmic beauty, and originality. This results in scores that fail to fully reflect the true quality of the essay. Furthermore, existing systems typically only output the overall score or grade of the entire essay, failing to provide granular evaluations and improvement suggestions for specific sentences, making it difficult for students to gain effective writing guidance from the scores. In addition, existing assessment methods generally use fixed scoring scales, failing to consider the differences in cognitive development levels and writing abilities among students of different grades. Scoring results may be too harsh for younger students and too lenient for older students, failing to provide targeted teaching incentives. Summary of the Invention

[0003] To address the aforementioned issues, this application provides an intelligent essay evaluation method and apparatus.

[0004] The first aspect of this application provides an intelligent essay evaluation method, characterized by comprising:

[0005] Step S1: Obtain the essay content displayed by the user in the scanning area using a scanning device;

[0006] Step S2: Evaluate each sentence of the composition at the lexical, semantic, prosodic, and novelty levels, and calculate the weighted score for each sentence.

[0007] Step S3: When the sentence score exceeds the set value, obtain the evaluation sentence evaluated by the large model in terms of vocabulary, semantics, prosody and novelty.

[0008] Step S4: Evaluate the essay's theme, structure, language, thinking, and norms based on the large model, and calculate the essay score using weighted averages.

[0009] Step S5: Adjust the essay score according to the following formula:

[0010] ;

[0011] in, The revised score. The score before correction. These are the boundary steepness control parameters. This is the full score for the essay. To get closer to a perfect score A given value, Scores that are correlated with the student's grade level;

[0012] Step S6: Project the corrected essay score onto the scanning area, and project the evaluation statement onto the side of the corresponding statement in the scanning area.

[0013] Preferably, step S1 further includes:

[0014] Based on the user's selection of the essay content, the selected content is used as the scanned content, while the unselected content is discarded; or based on the trained boundary recognition model, the boundaries of the essay content are identified, the content within the boundaries is used as the scanned content, and the content outside the boundaries is discarded.

[0015] Preferably, the score for the statement is calculated using the following formula:

[0016] ;

[0017] in, Score the vocabulary level. Score the semantic layer. Score the prosodic layer. Rate the novelty level. , , , These are the weighting coefficients.

[0018] Preferably, step S4 further includes:

[0019] The interface for setting the weighting parameters for each evaluation angle is presented to the user, and the weighting parameters for each evaluation angle given by the user according to the purpose of essay writing training are obtained.

[0020] The second aspect of this application provides an intelligent essay evaluation device, mainly comprising:

[0021] The essay content acquisition module is used to acquire the essay content displayed by the user in the scanning area through a scanning device;

[0022] The sentence scoring module is used to evaluate each sentence in the essay at the lexical, semantic, prosodic, and novelty levels, and to calculate the weighted score for each sentence.

[0023] The evaluation statement acquisition module is used to acquire the evaluation statement, which is evaluated by the large model in terms of vocabulary, semantics, prosody and novelty, when the score of the statement exceeds the set value.

[0024] The essay scoring module is used to evaluate essays based on a large model, considering aspects such as theme, structure, language, thinking, and norms, and then calculates the essay score using weighted averages.

[0025] The scoring correction module is used to correct essay scores according to the following formula:

[0026] ;

[0027] in, The revised score. The score before correction. These are the boundary steepness control parameters. This is the full score for the essay. To get closer to a perfect score A given value, Scores that are correlated with the student's grade level;

[0028] The content display module is used to project the corrected essay score onto the scanning area and project the evaluation statement onto the side of the corresponding statement in the scanning area.

[0029] Preferably, the essay content acquisition module includes:

[0030] The actual content determination unit is used to select the content in the frame of the essay content by the user and use the selected content as the scan content, while discarding the content that is not selected; or to identify the boundary of the essay content based on the trained boundary recognition model, use the content within the boundary as the scan content, and discard the content outside the boundary.

[0031] Preferably, in the statement score calculation module, the statement score is calculated using the following formula:

[0032] ;

[0033] in, Score the vocabulary level. Score the semantic layer. Score the prosodic layer. Rate the novelty level. , , , These are the weighting coefficients.

[0034] Preferably, the essay scoring module includes:

[0035] The parameter setting unit is used to present the scoring weighting parameter setting interface for each evaluation angle to the user and obtain the weighting parameters for each evaluation angle given by the user according to the purpose of essay training.

[0036] This application achieves refined evaluation of essays through multi-dimensional sentence evaluation and sentence-level large model feedback. Attached Figure Description

[0037] Figure 1 This is a flowchart of a preferred embodiment of the intelligent essay evaluation method of this application. Detailed Implementation

[0038] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions in the embodiments of this application will be described in more detail below with reference to the accompanying drawings. The described embodiments are only a part of the embodiments of this application, not all of them. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application. All other embodiments obtained by those skilled in the art based on the embodiments in this application without creative effort are within the scope of protection of this application. The embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0039] The first aspect of this application provides an intelligent essay evaluation method, such as... Figure 1 As shown, it mainly includes:

[0040] Step S1: Obtain the essay content displayed by the user in the scanning area using a scanning device;

[0041] Step S2: Evaluate each sentence of the composition at the lexical, semantic, prosodic, and novelty levels, and calculate the weighted score for each sentence.

[0042] Step S3: When the sentence score exceeds the set value, obtain the evaluation sentence evaluated by the large model in terms of vocabulary, semantics, prosody and novelty.

[0043] Step S4: Evaluate the essay's theme, structure, language, thinking, and norms based on the large model, and calculate the essay score using weighted averages.

[0044] Step S5: Adjust the essay score according to the following formula:

[0045] ;

[0046] in, The revised score. The score before correction. These are the boundary steepness control parameters. This is the full score for the essay. To get closer to a perfect score A given value, Scores that are correlated with the student's grade level;

[0047] Step S6: Project the corrected essay score onto the scanning area, and project the evaluation statement onto the side of the corresponding statement in the scanning area.

[0048] This application achieves refined evaluation of compositions through multi-dimensional sentence evaluation and sentence-level large model feedback, and introduces a grade-adapted scoring correction formula to make the scoring results more in line with the students' actual level.

[0049] This application obtains the user's essay content through step S1, preferably by scanning. The learning machine projects the essay onto the desktop. The projected content includes a control area and an input / output area. The control area mainly contains some operation buttons, and the input / output area is mainly used to scan the file placed by the user and project the content to be output to the user. Specifically, this application can place the essay part of the test paper in the scanning area, and then click the scan button in the control area to obtain the essay content.

[0050] Users of this application include students, teachers, or parents.

[0051] In some alternative implementations, step S1 further includes:

[0052] Based on the user's selection of the essay content, the selected content is used as the scanned content, while the unselected content is discarded; or based on the trained boundary recognition model, the boundaries of the essay content are identified, the content within the boundaries is used as the scanned content, and the content outside the boundaries is discarded.

[0053] This embodiment describes an adaptive setting of the scanning area through border recognition during a single scan. In practice, the learning machine is placed on a table, and it adaptively adjusts the initial scanning area position based on the distance between the scanning lens and the table. This position is typically adjusted during device installation. It's understood that after placing the exam paper in the initial scanning area, the essay content usually doesn't fill the entire area. Scanning the entire area would severely hinder the reading of the essay content, potentially including non-essay content like student names. This is especially true for exam papers where the essay occupies only a small portion. Therefore, after placing the exam paper in the initial scanning area, the user can mark specific scanning areas using gestures to select specific parts of the essay content. The device recognizes these selections and only reads that portion. Alternatively, the system can also adaptively set the final scanning area based on the exam paper's borders, simultaneously recognizing the user's intention to scan the essay.

[0054] Subsequently, in step S2, each sentence in the essay is scored.

[0055] In some alternative implementations, in step S2, the statement score is calculated using the following formula:

[0056] ;

[0057] in, Score the vocabulary level. Score the semantic layer. Score the prosodic layer. Rate the novelty level. , , , These are the weighting coefficients.

[0058] In this embodiment, among the four indicators for calculating the score of the above-mentioned statement, the lexical level score is used to measure the sophistication of word choice, the semantic level score is used to measure the depth of logic and artistic conception, the prosodic level score is used to measure the rhythmic beauty, and the novelty score is used to measure the innovation.

[0059] The formula for calculating the vocabulary level score is as follows:

[0060] ;

[0061] in, The words in the sentence are obtained by segmenting them using a word segmenter. For words Term frequency - inverse document frequency, For words The vocabulary elegance score is based on the word frequency classification in the *Modern Chinese Dictionary*, with a score ranging from 1 to 10. For sentences, The length of the sentence.

[0062] The formula for calculating the semantic layer score is:

[0063] ;

[0064] in, Used to describe semantic coherence, it is obtained through cosine similarity calculation using the BERT model; here it refers to the similarity between sentences. The cosine similarity is calculated between the sentence and the sentence before or after it. For example, if sentence A is "Last weekend, my dad and I went fishing by the river" and sentence B is "We sat for a long time and finally caught a big fish", both sentences mention fishing, so the similarity is high. The richness of imagery can be described using a rhetorical device detection model, such as a metaphor or personification detection model. This represents the weighting coefficient for rhetorical devices.

[0065] The formula for calculating the prosodic level score is:

[0066] ;

[0067] in, , As weight, , Sentence Chinese words The highest score for the pause rhythm can be obtained by matching the level and oblique patterns. The score is between 0 and 1. For example, the score for "level level oblique oblique" can be preset to 0.9. Sentence The rhyme intensity is determined by the ratio of the number of words in a sentence whose pauses exceed a set value to the total number of words in the sentence, with a score between 0 and 1.

[0068] The formula for calculating the novelty score is:

[0069] ;

[0070] in, A large model for calculating the similarity of terms, used to calculate sentences. With corpus The similarity of sentences in the corpus is calculated as a percentage. The corpus contains classic sentences from Tang and Song poetry as well as modern masterpieces.

[0071] By employing the above methods, this application refines essay evaluation from the traditional whole-essay scoring to the sentence level, and comprehensively evaluates each sentence simultaneously from four dimensions: vocabulary, semantics, rhythm, and originality. This overcomes the limitations of existing systems that rely solely on surface features or single-dimensional scoring. On one hand, the fine-grained sentence-level evaluation allows the system to accurately locate specific high-performing sentences in the essay, providing a basis for the targeted generation of sentence-level evaluation sentences from the large model in subsequent step S3. On the other hand, the four evaluation dimensions characterize sentence quality from different angles, such as word accuracy, sentence coherence, rhythm, and originality. The weighted and fused sentence scores more comprehensively and objectively reflect the true writing level of each sentence, avoiding the one-sidedness of single-indicator evaluation.

[0072] Subsequently, in step S3, the system backend can directly provide the above statement and evaluation prompts in terms of vocabulary, semantics, prosody and novelty to the large model, which will then directly generate the evaluation statement.

[0073] Next, in step S4, this application evaluates the essays using a large-scale model. The theme refers to the essay's content and central idea, examining the clarity of the theme, depth of thought, and quality of materials; structure refers to the essay's logical layout, examining logical levels, allocation of detail, and the design of the beginning and ending; language refers to the essay's expression, examining the accuracy, fluency, and expressiveness of language; thinking refers to the essay's creative highlights, examining the central idea, individuality, cultural relevance, critical thinking, and creativity; and standardization examines grammatical errors, misspellings, incorrect words, punctuation, word count, and format. It is understandable that the above-analyzed content can directly serve as guiding words / prompts for the large-scale model.

[0074] In an alternative implementation, this application can also independently train relevant models by manually evaluating the content of essays and labeling training samples from different evaluation perspectives, thereby training an independent model capable of examining the above four indicators.

[0075] In some alternative implementations, step S4 further includes:

[0076] The interface for setting the weighting parameters for each evaluation angle is presented to the user, and the weighting parameters for each evaluation angle given by the user according to the purpose of essay writing training are obtained.

[0077] For teachers, the grading criteria need to be adjusted accordingly based on the emphasis of different semesters or lesson plans. For example, if the main focus of this semester is on the application of rhetorical devices, the weighting parameter of the descriptive indicator of language can be increased.

[0078] Subsequently, in step S5, this application corrects the score given by the large model.

[0079] For example, full marks This score is the highest possible score that can be given for this essay. When manually grading students' essays, whether they are in first grade, second grade, or fifth grade, good essays usually receive a higher score, such as 8 or 9 points, based on the principle of fairness. However, for the large model, the essay grading algorithm is the same, and the scores of first-grade students are usually much lower than those of fifth-grade students. Therefore, it is necessary to adjust the scores given by the large model according to the overall learning level of different grades, for example, by increasing the essay scores of first-grade students.

[0080] Specifically, the teacher first determines the given value. For example, a good essay is considered to be rated 9 points, that is... Then, the highest score that the model can assign to a student's grade level essay is evaluated, which is the score related to the student's grade level. For example, for a first-grade student, the highest score the system might give is 3 points. According to the conventional algorithm, a first-grade student's essay score only needs to be multiplied by a factor. ( Substituting into By constructing a score that satisfies the conditions, we can obtain a score that meets the requirements. The linear correction function.

[0081] However, considering the scores mentioned above that are related to the student's grade level... This is a preliminary estimate; that is, it's possible that a first-grade student might write a very good essay, and the large model provides a score for that essay. More than 3 points, for example If we still multiply by a factor at this point The revised score is 12 points, which exceeds the maximum score. This is illogical, therefore this application discards the aforementioned linear function. Furthermore, to simplify the system model and prevent the introduction of excessive conditional statements, this embodiment ultimately constructs the following modified function:

[0082] .

[0083] In this embodiment, and The value can be stored in the system in advance and retrieved when necessary based on the student's grade level.

[0084] The purpose of the above correction function is to make the independent variable... When, dependent variable Able to increase rapidly In the independent variable When, dependent variable It can continue to slowly increase to a perfect score. And will not exceed the full score. .

[0085] Finally, in step S6, the essay scoring and evaluation statements are projected and displayed. Typically, after the essay content is acquired, the scanning area becomes the display area. Using a projection device, the content to be displayed can be directly superimposed onto the exam paper.

[0086] In an alternative implementation, when the evaluation statement is projected onto the side of the corresponding statement in the scanning area, the original statement is further annotated, for example by using underscores such as wavy lines or continuous circles, or by selecting the corresponding statement on the test paper with a square box.

[0087] This application achieves refined assessment of compositions through sentence-level multi-dimensional evaluation and grade-appropriate scoring correction, thereby improving the objectivity of the assessment and the relevance of teaching.

[0088] The second aspect of this application provides an intelligent essay evaluation device corresponding to the above method, mainly comprising:

[0089] The essay content acquisition module is used to acquire the essay content displayed by the user in the scanning area through a scanning device;

[0090] The sentence scoring module is used to evaluate each sentence in the essay at the lexical, semantic, prosodic, and novelty levels, and to calculate the weighted score for each sentence.

[0091] The evaluation statement acquisition module is used to acquire the evaluation statement, which is evaluated by the large model in terms of vocabulary, semantics, prosody and novelty, when the score of the statement exceeds the set value.

[0092] The essay scoring module is used to evaluate essays based on a large model, considering aspects such as theme, structure, language, thinking, and norms, and then calculates the essay score using weighted averages.

[0093] The scoring correction module is used to correct essay scores according to the following formula:

[0094] ;

[0095] in, The revised score. The score before correction. These are the boundary steepness control parameters. This is the full score for the essay. To get closer to a perfect score A given value, Scores that are correlated with the student's grade level;

[0096] The content display module is used to project the corrected essay score onto the scanning area and project the evaluation statement onto the side of the corresponding statement in the scanning area.

[0097] In some optional implementations, the essay content acquisition module includes:

[0098] The actual content determination unit is used to select the content in the frame of the essay content by the user and use the selected content as the scan content, while discarding the content that is not selected; or to identify the boundary of the essay content based on the trained boundary recognition model, use the content within the boundary as the scan content, and discard the content outside the boundary.

[0099] In some alternative implementations, the statement score is calculated in the statement score calculation module using the following formula:

[0100] ;

[0101] in, Score the vocabulary level. Score the semantic layer. Score the prosodic layer. Rate the novelty level. , , , These are the weighting coefficients.

[0102] In some optional implementations, the essay scoring module includes:

[0103] The parameter setting unit is used to present the scoring weighting parameter setting interface for each evaluation angle to the user and obtain the weighting parameters for each evaluation angle given by the user according to the purpose of essay training.

[0104] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. An intelligent essay evaluation method, characterized in that, include: Step S1: Obtain the essay content displayed by the user in the scanning area using a scanning device; Step S2: Evaluate each sentence of the composition at the lexical, semantic, prosodic, and novelty levels, and calculate the weighted score for each sentence. Step S3: When the sentence score exceeds the set value, obtain the evaluation sentence evaluated by the large model in terms of vocabulary, semantics, prosody and novelty. Step S4: Evaluate the essay's theme, structure, language, thinking, and norms based on the large model, and calculate the essay score using weighted averages. Step S5: Adjust the essay score according to the following formula: ； in, The revised score. The score before correction. These are the boundary steepness control parameters. This is the full score for the essay. To get closer to a perfect score A given value, Scores that are correlated with the student's grade level; Step S6: Project the corrected essay score onto the scanning area, and project the evaluation statement onto the side of the corresponding statement in the scanning area.

2. The intelligent essay evaluation method according to claim 1, characterized in that, Step S1 further includes: Based on the user's selection of the essay content, the selected content is used as the scanned content, while the unselected content is discarded; or based on the trained boundary recognition model, the boundaries of the essay content are identified, the content within the boundaries is used as the scanned content, and the content outside the boundaries is discarded.

3. The intelligent essay evaluation method according to claim 1, characterized in that, In step S2, the score for the statement is calculated using the following formula: ； in, Score the vocabulary level. Score the semantic layer. Score the prosodic layer. Rate the novelty level. , , , These are the weighting coefficients.

4. The intelligent essay evaluation method according to claim 1, characterized in that, Step S4 further includes: The interface for setting the weighting parameters for each evaluation angle is presented to the user, and the weighting parameters for each evaluation angle given by the user according to the purpose of essay writing training are obtained.

5. An intelligent essay evaluation device, characterized in that, include: The essay content acquisition module is used to acquire the essay content displayed by the user in the scanning area through a scanning device; The sentence scoring module is used to evaluate each sentence in the essay at the lexical, semantic, prosodic, and novelty levels, and to calculate the weighted score for each sentence. The evaluation statement acquisition module is used to acquire the evaluation statement, which is evaluated by the large model in terms of vocabulary, semantics, prosody and novelty, when the score of the statement exceeds the set value. The essay scoring module is used to evaluate essays based on a large model, considering aspects such as theme, structure, language, thinking, and norms, and then calculates the essay score using weighted averages. The scoring correction module is used to correct essay scores according to the following formula: ； in, This is the revised score. The score before correction. These are the boundary steepness control parameters. This is the full score for the essay. To get closer to a perfect score A given value, Scores that are correlated with the student's grade level; The content display module is used to project the corrected essay score onto the scanning area and project the evaluation statement onto the side of the corresponding statement in the scanning area.

6. The intelligent essay evaluation device according to claim 5, characterized in that, The essay content acquisition module includes: The actual content determination unit is used to select the content in the frame of the essay content by the user and use the selected content as the scan content, while discarding the content that is not selected; or to identify the boundary of the essay content based on the trained boundary recognition model, use the content within the boundary as the scan content, and discard the content outside the boundary.

7. The intelligent essay evaluation device according to claim 5, characterized in that, In the statement score calculation module, the statement score is calculated using the following formula: ； in, Score the vocabulary level. Score the semantic layer. Score the prosodic layer. Rate the novelty level. , , , These are the weighting coefficients.

8. The intelligent essay evaluation device according to claim 7, characterized in that, The essay scoring module includes: The parameter setting unit is used to present the scoring weighting parameter setting interface for each evaluation angle to the user and obtain the weighting parameters for each evaluation angle given by the user according to the purpose of essay training.

Images