Statement quality evaluation method based on rhetorical technique recognition and multi-feature fusion

Abstract

The invention provides a statement quality evaluation method based on rhetorical technique recognition and multi-feature fusion, and relates to the technical field of statement quality evaluation, and the method specifically comprises the steps: collecting sentence making data of pupils, and allocating part-of-speech tags for each word through part-of-speech tagging; performing feature extraction on nouns, verbs and adjectives of sentence making to obtain rhetorical features, including rhetorical techniques of sentence making and the number of rhetorical technique types; extracting basic feature parameters of past and current sentences, and training a scoring model through the basic features to obtain a basic quality coefficient; and combining the rhetorical features with the basic quality coefficient to generate a comprehensive scoring coefficient, and mapping the comprehensive scoring coefficient to an evaluation grade. According to the method, subjectivity and one-sidedness of a traditional evaluation method are avoided, the consistency and scientificity of scoring are improved, an education manager can conduct fair and just evaluation, a more specific guidance basis is provided for teachers, and the teachers are helped to make more effective teaching strategies.

Description

Technical Field

[0001] The present invention relates to the technical field of sentence quality evaluation, and in particular to a sentence quality evaluation method based on rhetoric recognition and multi-feature fusion. Background Art

[0002] Language and writing are important communication tools in all fields of human society. The core goal of Chinese courses is to guide students to practice language in a language application environment, and to experience the characteristics and application rules of language and writing. The training of writing ability reflects students' language expression ability, written expression ability and thinking organization ability. How to efficiently cultivate students' writing ability is an important topic to improve the quality of Chinese teaching. With the rapid development of artificial intelligence and information technology, the traditional way of correcting writing tasks has exposed many shortcomings: first, teachers need to spend more time and energy reading and correcting writing tasks, resulting in a long feedback cycle; second, teachers have different scoring habits for writing tasks, lack of unified scoring standards, and students often receive single feedback, which is difficult to meet the requirements of personalized guidance for students today. In order to solve the problems existing in the correction of writing tasks under the traditional model, the use of information technology artificial intelligence methods to automatically analyze and process writing data and construct an automatic evaluation model for writing task quality has become an important research direction in the development of educational informatization.

[0003] In the existing technology, the challenges faced by primary school students in language expression ability assessment are becoming more and more significant, especially the evaluation of sentence-making ability. This process often relies on the subjective judgment of teachers, resulting in inconsistent and unreliable scoring. Different teachers have different scoring standards for the same sentence, which not only affects students' learning confidence, but also makes it difficult for educational administrators to conduct fair and impartial evaluations. In addition, traditional evaluation methods usually fail to fully consider students' combination of rhetoric, vocabulary usage, logical connection and sentence structure. This limitation not only leads to the one-sidedness of the evaluation results, but also limits teachers' in-depth understanding of students' language ability.

[0004] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not constitute the prior art that is already known to one of ordinary skill in the art. Summary of the invention

[0005] The purpose of the present invention is to provide a sentence quality evaluation method based on rhetoric recognition and multi-feature fusion to solve the problems raised in the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A sentence quality evaluation method based on rhetoric recognition and multi-feature fusion, the specific steps include:

[0008] S1. Collect a dataset of primary school students’ sentences in the past, segment the sentences into words, assign a part-of-speech tag to each word in the sentence through part-of-speech tagging, and determine its role in the sentence, including noun, verb, and adjective;

[0009] S2. Extract features of nouns, verbs and adjectives in sentences to obtain rhetorical features, including rhetorical devices in sentences, the number of types of rhetorical devices and the number of each rhetorical device, mark rhetorical features on sentences and train rhetorical device recognition models, use the trained rhetorical device recognition model to recognize and classify rhetorical devices in sentences of current primary school students, and obtain the number of types of rhetorical devices and the number of each rhetorical device in sentences of current primary school students;

[0010] S3. extract the basic characteristic parameters of sentences made by primary school students in the past and at present, including the number of words in the sentence, the richness of vocabulary, the number of grammatical errors and the frequency of use of logical connectives, and set different coefficients for the number of words in the sentence, the richness of vocabulary, the number of grammatical errors and the frequency of use of logical connectives according to the numerical range;

[0011] S4. The number of characters, vocabulary richness, number of grammatical errors and frequency of use of logical connectives in the past sentences of primary school students after assignment are used as input, and the basic quality coefficient is used as the label output, and the sentence scoring model is trained to obtain the trained model, and the number of characters, vocabulary richness, number of grammatical errors and frequency of use of logical connectives in the current sentences of primary school students after assignment are input into the sentence scoring model to obtain the basic quality coefficient of the current sentences of primary school students;

[0012] S5. Perform data processing on the number of rhetorical types and the number of each rhetoric in the current primary school students' sentences to generate an advanced scoring coefficient for the current primary school students' sentences; perform data analysis and processing on the basic quality coefficient and the advanced scoring coefficient for the current primary school students' sentences to generate a comprehensive scoring coefficient for the current primary school students' sentences; map the comprehensive scoring coefficient and the sentence judgment level to obtain the judgment level of the current primary school students' sentences.

[0013] Furthermore, feature extraction is performed on the nouns, verbs and adjectives in the sentence to obtain rhetorical features, including the rhetorical devices in the sentence, the number of rhetorical device types and the number of each rhetorical device. The rhetorical features are annotated on the sentence and the rhetorical device recognition model is trained. The specific process is as follows:

[0014] Extract nouns, verbs, and adjectives from sentence formation;

[0015] Identify the rhetorical devices that need to be identified, including metaphor, personification, and parallelism;

[0016] Count the number of rhetorical device types and the number of each rhetorical device;

[0017] Use word embedding technology to calculate the similarity between words;

[0018] Extract the sentence's syntactic tree structure and analyze the syntactic features of rhetoric;

[0019] Design category labels for each rhetorical device for annotation;

[0020] According to the extracted features and defined rhetorical devices, the sentences are annotated, and the rhetorical devices and their number in each sentence are annotated in the sentence;

[0021] The annotated sentences are used as training data sets to construct feature matrices, with rhetorical devices and their numbers as labels;

[0022] Choose a suitable machine learning model;

[0023] Use the training data set to train the model, and the model needs to be able to identify different rhetorical devices in the sentence;

[0024] Apply the trained model to new sentences to identify their rhetorical features;

[0025] The trained rhetoric device recognition model is used to identify and classify the rhetoric devices in the current primary school students' sentences, and the types and quantities of rhetoric devices in the current primary school students' sentences are obtained. The specific process is as follows:

[0026] Input the current primary school student sentence into the trained rhetorical device recognition model, and use the previously learned features to identify the rhetorical devices in the sentence, as follows:

[0027] Feature extraction: extract part of speech and context information from the current sentence;

[0028] Classification: Label the identified parts of the sentence as specific rhetorical devices based on the extracted parts of speech and contextual information;

[0029] The rhetorical device recognition model will output the type and number of rhetorical devices recognized in the current sentence, "metaphor" or "personification", and record the number of times they appear respectively.

[0030] Furthermore, the vocabulary richness and the frequency of use of logical connectives are obtained according to the following formula:

[0031]

[0032] Among them, CHD is the vocabulary richness, ZCH is the total vocabulary, CYS is the total number of words, PL is the frequency of use of logical connectives, and N is the number of times the logical connectives are used.

[0033] Furthermore, the sentence scoring model is composed of a deep neural network based on a multilayer perceptron, wherein the deep neural network of the multilayer perceptron includes an input layer, a first hidden layer, a second hidden layer, a third hidden layer and an output layer, wherein the first hidden layer, the second hidden layer and the third hidden layer each have at least two neurons, and each uses ReLU as an activation function;

[0034] The process of training the sentence scoring model is as follows:

[0035] Based on the number of words, vocabulary richness, number of grammatical errors and frequency of use of logical conjunctions in the primary school students' past sentences after assignment, an expert group evaluated the corresponding basic quality coefficient. The value range of the basic quality coefficient is between 0 and 10. The larger the basic quality coefficient, the better the basic quality of the primary school students' sentence construction.

[0036] Furthermore, the number of types of rhetorical devices and the number of each rhetorical device in the current primary school students' sentences are processed to generate the advanced scoring coefficient of the current primary school students' sentences, based on the following formula:

[0037]

[0038] Among them, JP is the current advanced scoring coefficient of primary school students' sentence making, R i is the number of the i-th rhetorical device, i is the index of the rhetorical device, and the value range of i is [1, m]. max is the preset value of rhetoric, m is the number of rhetoric types;

[0039] The value range of the advanced scoring coefficient JP is [0,1];

[0040] JP=0: indicates that no rhetoric devices are used in the sentence, indicating that the student's sentence-making ability is very weak in terms of rhetoric application and lacks diversity and creativity;

[0041] JP=1: It means that the maximum number of all rhetorical devices used in the sentence has been reached, indicating that the student can skillfully and richly use various rhetorical devices in sentence construction, showing a high level of language expression ability and creativity.

[0042] Furthermore, the basic quality coefficient and advanced scoring coefficient of the current primary school students' sentence making are analyzed and processed to generate the comprehensive scoring coefficient of the current primary school students' sentence making, based on the following formula:

[0043] ZP=Q c JP k

[0044] Among them, ZP is the comprehensive scoring coefficient of the current primary school students' sentence making, Q cQ is the basic quality coefficient of the current primary school students' sentence-making, and k is the exponent that controls the influence degree of the advanced scoring coefficient on the comprehensive scoring coefficient;

[0045] The value range of the advanced scoring coefficient ZP is [0, 10];

[0046] 0 < ZP ≤ 3: The quality of the students' sentence-making needs to be significantly improved, and the basic quality coefficient Q c is low or the advanced scoring coefficient JP is significantly insufficient;

[0047] 3 < ZP ≤ 6: The students have a certain foundation in sentence-making ability, but there is still room for improvement. The basic quality and advanced scoring are combined, but not excellent;

[0048] 6 < ZP ≤ 8: The students have strong sentence-making ability, possess certain expression skills and creativity, and the basic quality and advanced scoring are high;

[0049] 8 < ZP < 10: The students' sentence-making ability is very excellent, and they can flexibly use a variety of language skills, with clear expression and rich creativity.

[0050] Furthermore, map the comprehensive scoring coefficient and the sentence-making judgment level to obtain the judgment level of the current primary school students' sentence-making. The specific process is as follows:

[0051] When 0 < ZP ≤ 3: The judgment level is poor;

[0052] When 3 < ZP ≤ 6: The judgment level is medium;

[0053] When 6 < ZP ≤ 8: The judgment level is good;

[0054] When 8 < ZP < 10: The judgment level is excellent.

[0055] Compared with the prior art, the beneficial effects of the present invention are:

[0056] Firstly, the present invention uses the part-of-speech tagging technology to conduct detailed word segmentation and part-of-speech marking on the primary school students' sentence-making, clarifying the role of each word in the sentence. This greatly enhances the understanding of the sentence structure and lays a foundation for the subsequent extraction of rhetorical devices; secondly, using the trained rhetorical device recognition model, the system identifies the rhetorical features in the sentence, including metaphor, personification, and parallelism. This measure not only enriches the evaluation dimension but also provides a quantitative evaluation of the students' language innovation ability. At the same time, by extracting basic features such as the number of characters in the sentence, lexical richness, the number of grammar errors, and the usage frequency of logical connectives, a multi-dimensional scoring system is constructed. This method ensures the comprehensiveness of the evaluation and enables different levels of language abilities to be fully considered;

[0057] By integrating historical and current data to form a dynamic evaluation mechanism, the subjectivity and one-sidedness of traditional evaluation methods are avoided. The introduction of a comprehensive scoring model allows students to not only see their performance in basic language skills, but also gain a deep understanding and feedback on their expression skills through the identification of rhetorical techniques. Such an evaluation not only improves the consistency and scientific nature of the scoring, allowing educational administrators to conduct fair and just evaluations, but also provides teachers with more specific guidance to help them develop more effective teaching strategies. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 It is a schematic diagram of the overall method flow of the present invention. DETAILED DESCRIPTION

[0059] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with specific embodiments.

[0060] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the present invention should be understood by people with ordinary skills in the field to which the present invention belongs. The words "first", "second" and similar words used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0061] Embodiment 1:

[0062] See also Figure 1 , the present invention provides a technical solution:

[0063] A sentence quality evaluation method based on rhetoric recognition and multi-feature fusion, the specific steps include:

[0064] S1. Collect a dataset of primary school students’ sentences in the past, segment the sentences into words, assign a part-of-speech tag to each word in the sentence through part-of-speech tagging, and determine its role in the sentence, including noun, verb, and adjective;

[0065] S2. Extract features of nouns, verbs and adjectives in sentences to obtain rhetorical features, including rhetorical devices in sentences, the number of types of rhetorical devices and the number of each rhetorical device, mark rhetorical features on sentences and train rhetorical device recognition models, use the trained rhetorical device recognition model to recognize and classify rhetorical devices in sentences of current primary school students, and obtain the number of types of rhetorical devices and the number of each rhetorical device in sentences of current primary school students;

[0066] S3. extract the basic characteristic parameters of sentences made by primary school students in the past and at present, including the number of words in the sentence, the richness of vocabulary, the number of grammatical errors and the frequency of use of logical connectives, and set different coefficients for the number of words in the sentence, the richness of vocabulary, the number of grammatical errors and the frequency of use of logical connectives according to the numerical range;

[0067] S4. The number of characters, vocabulary richness, number of grammatical errors and frequency of use of logical connectives in the past sentences of primary school students after assignment are used as input, and the basic quality coefficient is used as the label output, and the sentence scoring model is trained to obtain the trained model, and the number of characters, vocabulary richness, number of grammatical errors and frequency of use of logical connectives in the current sentences of primary school students after assignment are input into the sentence scoring model to obtain the basic quality coefficient of the current sentences of primary school students;

[0068] S5. Perform data processing on the number of rhetorical types and the number of each rhetoric in the current primary school students' sentences to generate an advanced scoring coefficient for the current primary school students' sentences; perform data analysis and processing on the basic quality coefficient and the advanced scoring coefficient for the current primary school students' sentences to generate a comprehensive scoring coefficient for the current primary school students' sentences; map the comprehensive scoring coefficient and the sentence judgment level to obtain the judgment level of the current primary school students' sentences.

[0069] Based on the above embodiment, the specific process of step S1 is as follows:

[0070] There are many ways to collect sentence-making data from primary school students, such as asking students to complete sentence-making exercises in class, collecting their sentences through questionnaires, or obtaining them in daily writing, ensuring that the collected sentences are diverse for subsequent analysis;

[0071] After collecting the data, data preprocessing was performed. This step mainly included: removing punctuation marks, blank characters, and other non-text elements in the sentences;

[0072] Segment the processed sentences. Segmentation is the process of dividing sentences into independent words. You can use the segmentation tool to extract the words in the sentence. For example, after the sentence "I like kittens" is segmented, the words obtained are "I", "like", and "kittens";

[0073] After getting the words, assign a part-of-speech tag to each word. Part-of-speech tagging is the process of determining the grammatical role of each word in a sentence. This can be achieved by using a part-of-speech tagging tool. Each word will be marked with the corresponding part of speech, such as noun, verb, adjective, etc. For example, "kitten" is marked as a noun and "like" is marked as a verb.

[0074] By analyzing the results of part-of-speech tagging, we can determine the specific role of each word in the sentence and classify the part of speech of each word, for example:

[0075] Noun: refers to a person, thing or place;

[0076] Verb: expresses action or state;

[0077] Adjective: describes the characteristics or qualities of a noun.

[0078] On the basis of the above embodiment, feature extraction is performed on the nouns, verbs and adjectives in the sentence to obtain rhetorical features, including the rhetorical devices in the sentence, the number of rhetorical device types and the number of each rhetorical device. The rhetorical features are marked on the sentence and the rhetorical device recognition model is trained. The specific process is as follows:

[0079] Extract nouns, verbs, and adjectives from sentence formation;

[0080] Identify the rhetorical devices that need to be identified, including metaphor, personification, and parallelism;

[0081] Count the number of rhetorical device types and the number of each rhetorical device;

[0082] Use word embedding technology (such as Word2Vec, GloVe) to calculate the similarity between words;

[0083] Extract the sentence's syntactic tree structure and analyze the syntactic features of rhetoric;

[0084] Design a labeling system for each rhetorical device, which can be a simple category label (such as "metaphor");

[0085] According to the extracted features and defined rhetorical devices, the sentences are annotated. The rhetorical devices and their number in each sentence can be annotated in the sentence.

[0086] The annotated sentences are used as training data sets to construct feature matrices, and rhetorical devices and their numbers can be used as labels;

[0087] Choose a suitable machine learning model;

[0088] Use the training data set to train the model, and the model needs to be able to identify different rhetorical devices in the sentence;

[0089] Apply the trained model to new sentences to identify their rhetorical features.

[0090] The trained rhetoric device recognition model is used to identify and classify the rhetoric devices in the current primary school students' sentences, and the types and quantities of rhetoric devices in the current primary school students' sentences are obtained. The specific process is as follows:

[0091] Input the current primary school student's sentence into the trained rhetorical device recognition model. The model will analyze the sentence and use the previously learned features to identify the rhetorical devices in the sentence. The specific steps include:

[0092] Feature extraction: extract part of speech and context information from the current sentence;

[0093] Classification: Label the identified parts of the sentence as specific rhetorical devices based on the extracted parts of speech and contextual information;

[0094] The model will output the type and number of rhetorical devices recognized in the current sentence. For example, it will recognize "metaphor" and "personification" and record the number of times they appear respectively.

[0095] Based on the above embodiment, the vocabulary richness and the frequency of use of logical connectives are obtained according to the following formula:

[0096]

[0097] Among them, CHD is the vocabulary richness, ZCH is the total vocabulary, CYS is the total number of words, PL is the frequency of use of logical connectives, and N is the number of times the logical connectives are used.

[0098] On the basis of the above embodiment, different coefficients are assigned to sentence length, vocabulary richness, number of grammatical errors and frequency of use of logical connectives according to numerical ranges. The specific process is as follows:

[0099] Number of words in a sentence:

[0100] 1-5 words: coefficient is 1; 6-10 words: coefficient is 2; 11 words and above: coefficient is 3.

[0101] Vocabulary Richness:

[0102] 0-0.2: the coefficient is 1; 0.21-0.5: the coefficient is 2; 0.51-1.0: the coefficient is 3.

[0103] Number of syntax errors:

[0104] 0 errors: factor 3; 1-2 errors: factor 2; 3 errors and above: factor 1.

[0105] For example, there are two grammatical errors in the sentence "They are students and teachers". The first "teacher's" should be changed to "teachers" and the second "students and teachers" should be changed to "both students and teachers".

[0106] Frequency of use of logical connectives:

[0107] 0-0.1: the coefficient is 1; 0.11-0.3: the coefficient is 2; 0.31 and above: the coefficient is 3.

[0108] Based on the above embodiment, the sentence scoring model is composed of a deep neural network based on a multilayer perceptron, and the deep neural network of the multilayer perceptron includes an input layer, a first hidden layer, a second hidden layer, a third hidden layer and an output layer. The first hidden layer, the second hidden layer and the third hidden layer each have at least two neurons, and each uses ReLU as an activation function.

[0109] In this embodiment, the input features of the deep neural network of the multilayer perceptron include four features: the number of words in the sentence after the coefficient is assigned, the vocabulary richness, the number of grammatical errors and the frequency of use of logical connectives.

[0110] The structure of a deep neural network with a multilayer perceptron is:

[0111] Input layer: receives input of 4 features;

[0112] The first hidden layer has 64 neurons and uses ReLU as the activation function.

[0113] The second hidden layer has 128 neurons and also uses the ReLU activation function.

[0114] The third hidden layer has 64 neurons and uses the ReLU activation function.

[0115] Output layer: has a single neuron, the basic quality coefficient of the current primary school student's sentence construction.

[0116] The process of training the sentence scoring model is as follows:

[0117] Based on the number of words, vocabulary richness, number of grammatical errors and frequency of use of logical conjunctions in the primary school students' past sentences after assignment, an expert group evaluated the corresponding basic quality coefficient. The value range of the basic quality coefficient is between 0 and 10. The larger the basic quality coefficient, the better the basic quality of the primary school students' sentence construction.

[0118] The number of words, vocabulary richness, number of grammatical errors and frequency of use of logical connectives in the assigned sentences are used as input, and the corresponding basic quality coefficients are used as output labels for training. The mean square error is used as the loss function. When the mean square error is within [0, 0.1], the training of the sentence scoring model is completed.

[0119] The reason why the value range of the basic quality coefficient is controlled between 0 and 10 is as follows:

[0120] The range of 0 to 10 makes grading more intuitive. Students, teachers, and parents can easily understand and accept such a simple grading system. In contrast, the range of 0 to 1 may give people a vague feeling, and 0 to 100 may seem too complicated.

[0121] The 0 to 10 range provides enough granularity to effectively distinguish different levels of sentence-making ability. For example, each score in the 0 to 10 range can represent a different level without causing confusion due to too many subdivisions. The 0 to 1 subdivision may seem too coarse, while the 0 to 100 scale may be too detailed to be managed.

[0122] On a scale of 0 to 10, students can more easily perceive progress. When they move from a 5 to a 6, they can more intuitively feel that their efforts have paid off. If a scale of 0 to 1 is used, the change will seem insignificant (such as from 0.5 to 0.6), and 0 to 100 may make students feel that the goal is too far away.

[0123] Using a 0 to 10 scoring model makes it easy to perform statistical analysis, such as calculating the mean score, standard deviation, etc., without having to consider more complex proportional conversions.

[0124] On the basis of the above embodiment, the number of types of rhetorical devices and the number of each rhetorical device in the current primary school students' sentences are processed to generate the advanced scoring coefficient of the current primary school students' sentences, according to the following formula:

[0125]

[0126] Among them, JP is the current advanced scoring coefficient of primary school students' sentence making, R i is the number of the i-th rhetorical device, i is the index of the rhetorical device, and the value range of i is [1, m]. max is the preset value of the rhetorical device, and m is the number of rhetorical device types.

[0127] The value range of the advanced scoring coefficient JP is [0,1].

[0128] JP=0: indicates that no rhetoric devices are used in the sentence, indicating that the student's sentence-making ability is very weak in terms of rhetoric application and lacks diversity and creativity;

[0129] JP=1: It means that the maximum number of all rhetorical devices used in the sentence has been reached, indicating that the student can skillfully and richly use various rhetorical devices in sentence construction, showing a high level of language expression ability and creativity.

[0130] R max It can be preset according to the course requirements and teaching objectives. For example, if the teaching objective is to enable students to flexibly use various rhetorical devices in sentences, a relatively high R max (such as 5 or 6) can be set to encourage students to use them diversely;

[0131] Set a reasonable maximum value according to the students' grades and writing abilities. For lower-grade primary school students, it may be set to 3 or 4; for higher-grade students, it can be set to 5 or higher.

[0132] Based on the above embodiments, data analysis and processing are performed on the basic quality coefficient and advanced scoring coefficient of current primary school students' sentence-making to generate the comprehensive scoring coefficient of current primary school students' sentence-making. The formula is as follows:

[0133] ZP = Q c ·JP k

[0134] Among them, ZP is the comprehensive scoring coefficient of current primary school students' sentence-making, Q c is the basic quality coefficient of current primary school students' sentence-making, and k is an exponent that controls the influence degree of the advanced scoring coefficient on the comprehensive scoring coefficient.

[0135] The value range of the advanced scoring coefficient ZP is [0, 10].

[0136] 0 < ZP ≤ 3: The quality of the students' sentence-making needs to be significantly improved. It may be that the basic quality coefficient Q c is relatively low, or the advanced scoring coefficient JP is significantly insufficient;

[0137] 3 < ZP ≤ 6: The students have a certain foundation in sentence-making ability, but there is still room for improvement. The basic quality and advanced scoring are combined, but not excellent;

[0138] 6 < ZP ≤ 8: The students have strong sentence-making ability, possess certain expression skills and creativity, and both the basic quality and advanced scoring are relatively high;

[0139] 8 < ZP < 10: The students' sentence-making ability is very excellent, and they can flexibly use various language skills, with clear expression and rich creativity.

[0140] Q c represents the basic ability of primary school students in sentence-making, such as sentence structure, grammar use, and basic vocabulary, etc. It is the basis of the comprehensive scoring. Only by laying a good foundation can higher-quality expressions be made. In the comprehensive scoring, since the value range of JP is [0, 1], even if the score of JP is very high, if Q cIf the score is low, the overall ZP will also be significantly affected, indicating that the basic quality is the foundation of students' language ability.

[0141] Q c As a linear coefficient, it means that regardless of the impact of the advanced score, Q c will directly affect the final comprehensive score. For students with poor basic abilities, even if they perform well in the advanced score, the comprehensive score coefficient ZP cannot reach a high level.

[0142] JP measures the ability of students to use rhetorical devices in sentence construction, such as metaphors, personifications, and parallelisms. These advanced abilities can reflect students' creativity and expression ability. The influence of JP depends on Q c 's basic score. Only when Q c reaches a certain standard can the improvement of JP have a significant positive impact on the comprehensive score coefficient ZP.

[0143] By introducing the exponent k, a non-linear amplification or reduction effect on the advanced score coefficient JP is achieved. Different k values can flexibly control the impact of the advanced score coefficient JP on the comprehensive score.

[0144] When k > 1: An increase in the advanced score coefficient JP will significantly amplify the result of the comprehensive score, which is suitable for emphasizing language expression ability and creativity;

[0145] When k < 1: It means that the influence of the advanced score coefficient JP is weakened, emphasizing the importance of basic quality. It is suitable for students with poor basic quality to avoid over-evaluating due to a one-time advanced performance. For those students with weak basic abilities, if they rely solely on a one-time advanced performance (such as using several rhetorical devices), it may lead to an overly high comprehensive score, thus not truly reflecting the overall ability of the students.

[0146] In summary, by constructing the above function expression of the comprehensive score coefficient, it can effectively reflect and evaluate the relationship among the basic quality, advanced ability, and comprehensive ability of primary school students in the process of sentence construction, reasonably and appropriately evaluate the current sentence-making ability of primary school students, which can well fit the actual situation and adapt to the needs of different students, promoting the all-round development of their language ability. This method not only provides a flexible evaluation tool for teachers but also provides a clear improvement direction for students, thus having high practical application value in teaching practice.

[0147] Based on the above embodiments, map the comprehensive score coefficient and the sentence-making judgment level to obtain the sentence-making judgment level of the current primary school students. The specific process is as follows:

[0148] When 0 < ZP ≤ 3: The judgment level is poor;

[0149] When 3 < ZP ≤ 6: The evaluation level is medium;

[0150] When 6 < ZP ≤ 8: The evaluation level is good;

[0151] When 8 < ZP < 10: The evaluation level is excellent.

[0152] The above formulas are all dimensionless and take their numerical values for calculation. The formula is obtained by collecting a large amount of data for software simulation to get a formula closest to the actual situation. The preset parameters in the formula are set by those skilled in the art according to the actual situation.

[0153] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. Those skilled in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed in this article can be implemented by, electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed by hardware or software methods depends on the specific application and design constraints of the technical solution.

[0154] The units described as separate components may or may not be physically separated. The components shown as units may or may not be physical units. They can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0155] The above is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art within the technical scope disclosed in this application can easily think of changes or substitutions, which should all be covered within the protection scope of this application.

[0156] The units described as separate components may or may not be physically separated. The components shown as units may or may not be physical units. They can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0157] The above is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art within the technical scope disclosed in this application can easily think of changes or substitutions, which should all be covered within the protection scope of this application.

Claims

1. A sentence quality evaluation method based on rhetoric recognition and multi-feature fusion, characterized in that: The specific steps include: S1. Collect a dataset of primary school students’ sentences in the past, segment the sentences into words, assign a part-of-speech tag to each word in the sentence through part-of-speech tagging, and determine its role in the sentence, including noun, verb, and adjective; S2. Extract features of nouns, verbs and adjectives in sentences to obtain rhetorical features, including rhetorical devices in sentences, the number of types of rhetorical devices and the number of each rhetorical device, mark rhetorical features on sentences and train rhetorical device recognition models, use the trained rhetorical device recognition model to recognize and classify rhetorical devices in sentences of current primary school students, and obtain the number of types of rhetorical devices and the number of each rhetorical device in sentences of current primary school students; S3. extract the basic characteristic parameters of sentences made by primary school students in the past and at present, including the number of words in the sentence, the richness of vocabulary, the number of grammatical errors and the frequency of use of logical connectives, and assign different coefficients to the number of words in the sentence, the richness of vocabulary, the number of grammatical errors and the frequency of use of logical connectives according to the numerical range; S4. The number of characters, vocabulary richness, number of grammatical errors and frequency of use of logical connectives in the past sentences of primary school students after assignment are used as input, and the basic quality coefficient is used as the label output, and the sentence scoring model is trained to obtain the trained model, and the number of characters, vocabulary richness, number of grammatical errors and frequency of use of logical connectives in the current sentences of primary school students after assignment are input into the sentence scoring model to obtain the basic quality coefficient of the current sentences of primary school students; S5. Perform data processing on the number of rhetorical types and the number of each rhetoric in the current primary school students' sentences to generate an advanced scoring coefficient for the current primary school students' sentences; perform data analysis and processing on the basic quality coefficient and the advanced scoring coefficient for the current primary school students' sentences to generate a comprehensive scoring coefficient for the current primary school students' sentences; map the comprehensive scoring coefficient and the sentence judgment level to obtain the judgment level of the current primary school students' sentences.

2. The sentence quality evaluation method based on rhetoric recognition and multi-feature fusion according to claim 1 is characterized in that: Feature extraction is performed on the nouns, verbs, and adjectives in the sentence to obtain rhetorical features, including the rhetorical devices in the sentence, the number of rhetorical device types, and the number of each rhetorical device. The rhetorical features are marked on the sentence and the rhetorical device recognition model is trained. The specific process is as follows: Extract nouns, verbs, and adjectives from sentence formation; Identify the rhetorical devices that need to be identified, including metaphor, personification, and parallelism; Count the number of rhetorical device types and the number of each rhetorical device; Use word embedding technology to calculate the similarity between words; Extract the sentence's syntactic tree structure and analyze the syntactic features of rhetoric; Design category labels for each rhetorical device for annotation; According to the extracted features and defined rhetorical devices, the sentences are annotated, and the rhetorical devices and their number in each sentence are annotated in the sentence; The annotated sentences are used as training data sets to construct feature matrices, with rhetorical devices and their numbers as labels; Choose a suitable machine learning model; Train the model using the training dataset; Apply the trained model to new sentences to identify their rhetorical features; Use the trained rhetorical device recognition model to identify and classify the rhetorical devices in the current sentences made by primary school students, and obtain the types and quantities of rhetorical devices in the current sentences made by primary school students. The specific process is as follows: Input the current sentences made by primary school students into the trained rhetorical device recognition model, and use the previously learned features to identify the rhetorical devices in the sentences, as follows: Feature extraction: Extract the part-of-speech and context information from the current sentence; Classification: According to the extracted part-of-speech and context information, label the identified parts in the sentence as specific rhetorical devices; The rhetorical device recognition model will output the types and quantities of rhetorical devices recognized in the current sentence, such as "metaphor" or "personification", and record the number of times each appears respectively.

3. The sentence quality evaluation method based on rhetoric recognition and multi-feature fusion according to claim 2 is characterized in that: Obtain the lexical richness and the usage frequency of logical connectives, based on the following formula: Among them, CHD is the lexical richness, ZCH is the total number of words, CYS is the total number of words, PL is the usage frequency of logical connectives, and N is the number of times logical connectives are used.

4. The sentence quality evaluation method based on rhetoric recognition and multi-feature fusion according to claim 3 is characterized by: The sentence scoring model is composed of a deep neural network based on a multi-layer perceptron. The deep neural network of the multi-layer perceptron includes an input layer, a first hidden layer, a second hidden layer, a third hidden layer, and an output layer. The first hidden layer, the second hidden layer, and the third hidden layer all have at least two neurons, and all use ReLU as the activation function; The process of training the sentence scoring model is as follows: According to the number of characters, lexical richness, the number of grammar errors, and the usage frequency of logical connectives in the past sentences made by primary school students after assignment, the expert group evaluates the corresponding basic quality coefficient. The value range of the basic quality coefficient is between 0 and 10. The larger the basic quality coefficient, the better the basic quality of the sentences made by primary school students.

5. The sentence quality evaluation method based on rhetoric recognition and multi-feature fusion according to claim 4 is characterized in that: Process the quantity of the types of rhetorical devices in the current sentences made by primary school students and the quantity of each rhetorical device to generate the advanced scoring coefficient of the current sentences made by primary school students, based on the following formula: Among them, JP is the current advanced scoring coefficient of primary school students' sentence making, R i is the number of the i-th rhetorical device, i is the index of the rhetorical device, and the value range of i is [1, m]. max is the preset value of rhetoric, m is the number of rhetoric types; The value range of the advanced scoring coefficient JP is [0, 1]; JP = 0: It means that no rhetorical devices are used in the sentence, indicating that the student's sentence-making ability is very weak in the application of rhetoric, lacking diversity and creativity; JP = 1: It means that all the rhetorical devices used in the sentence reach the maximum quantity, indicating that the student can proficiently and richly use various rhetorical devices in sentence-making, showing a high language expression ability and creativity.

6. The sentence quality evaluation method based on rhetoric recognition and multi-feature fusion according to claim 5 is characterized in that: Perform data analysis and processing on the basic quality coefficient and the advanced scoring coefficient of the current sentences made by primary school students to generate the comprehensive scoring coefficient of the current sentences made by primary school students, based on the following formula: ZP=Q c ·JP k Among them, ZP is the comprehensive scoring coefficient of the current primary school students' sentence making, Q c is the basic quality coefficient of the current primary school students' sentence making, k is an index, which controls the influence of the advanced scoring coefficient on the comprehensive scoring coefficient; The value range of the advanced scoring coefficient ZP is [0, 10]; 0 < ZP ≤ 3: The sentence-making quality of the student needs to be significantly improved, with the basic quality coefficient Q c being low or the advanced scoring coefficient JP being significantly insufficient; 3 < ZP ≤ 6: The student's sentence-making ability has a certain foundation, but there is still room for improvement. The basic quality and the advanced scoring are combined, but not excellent; 6 < ZP ≤ 8: The student's sentence-making ability is strong, with certain expression skills and creativity, and the basic quality and the advanced scoring are high; 8 < ZP < 10: The student's sentence-making ability is very excellent, able to flexibly use a variety of language skills, with clear expression and rich creativity.

7. The sentence quality evaluation method based on rhetoric recognition and multi-feature fusion according to claim 6 is characterized in that: Map the comprehensive scoring coefficient and the sentence judgment level to obtain the judgment level of the current sentences made by primary school students. The specific process is as follows: When 0 < ZP ≤ 3: The evaluation grade is poor; When 3 < ZP ≤ 6: The evaluation grade is medium; When 6 < ZP ≤ 8: The evaluation grade is good; When 8 < ZP < 10: The evaluation grade is excellent.

Images