Method and system for rapidly extracting metaphor in English evaluation process based on large language model

Abstract

The invention discloses a method and a system for realizing rapid extraction of metaphors in an English evaluation process based on a large language model, and aims to improve the efficiency and the accuracy of metaphor identification and provide accurate technical support for English teaching and language understanding. The method comprises the following steps: receiving multi-modal input and performing grammar processing to obtain lexical, syntactic and semantic data; then, performing feature extraction on the data by using a large language model, and quickly identifying and extracting metaphor expressions through matching of context information and metaphor rules; the system comprises a data preprocessing module, a data training module, a reasoning analysis module and a metaphor data set output module, the metaphor extraction process is optimized by adopting a Transform architecture, and the model precision is improved through machine learning.

Description

Technical Field

[0001] The present invention belongs to the technical field of natural language processing, and in particular relates to a method and system for quickly extracting metaphors in an English assessment process based on a large language model. Background Art

[0002] In traditional English learning and assessment, the ability to understand and use metaphors is one of the important indicators to measure students' language proficiency. However, existing metaphor recognition methods mainly rely on manual analysis or simple rule matching, which are not only inefficient but also have limited success rates, making it difficult to meet the needs of large-scale teaching and assessment. Summary of the invention

[0003] To solve the above problems, the present invention discloses a method and system for quickly extracting metaphors in the English assessment process based on a large language model. In addition to the traditional metaphor extraction scheme in the English assessment process, the present invention provides a new extraction and evaluation method to quickly and automatically extract metaphors, reduce labor input costs, and increase evaluation efficiency.

[0004] To achieve the above object, the technical solution of the present invention is as follows: A method for quickly extracting metaphors in an English assessment process based on a large language model comprises the following steps: S1. Process the evaluation request to form lexical data, syntactic data and semantic understanding data; S2. Extract features from the lexical data, syntactic data, and semantic understanding data processed by S1 to form context information. By matching the context information with metaphor rules, identify metaphorical expressions in the English assessment request. According to the metaphor rules, identify whether there are metaphors between sentences and output metaphor results. S3. Match the metaphor result output by S2 with the metaphor rule set. If the metaphor match succeeds, manually review the metaphor evaluation result and give a score. If the metaphor match fails, proceed to S4. S4. Metaphors are automatically generated for incorrect outputs through metaphor datasets. First, the resulting metaphor dataset after feature extraction is compared to match metaphor rules and generate lexical and syntactic sentences that do not meet the conditions. The metaphor dataset generates modification suggestions based on metaphor rules. After executing the suggestions, a sentence that conforms to the meaning of the original sentence and contains metaphors is generated.

[0005] As an improvement of the present invention, step S1 specifically includes the following contents: The evaluation request content is multimodal and can be text, voice, or video; After the evaluation request is processed, the lexical data is obtained through lexical analysis to form word segmentation results and part-of-speech tagging results. The syntactic data is obtained through syntactic analysis to obtain grammatical structure recognition results and short sentence recognition results. The semantic understanding data is obtained through semantic understanding, and a lexical semantic analysis result and a sentence semantic understanding result are obtained.

[0006] As an improvement of the present invention, the lexical analysis adopts a rule-based word segmentation algorithm and a statistics-based word segmentation algorithm, wherein: The rule-based word segmentation algorithm adopts the forward maximum matching method to match English words, taking a number of characters in the English corpus to be segmented from left to right to form a candidate word substring. If the substring matches a word in the vocabulary, it is segmented as a word, and this process is repeated until the entire corpus is segmented; The statistical-based word segmentation algorithm adopts the N-gram language model, performs word segmentation by counting the frequencies of N consecutive words in an English corpus, and counts the probabilities of multiple consecutive words appearing, thereby determining the best word segmentation position.

[0007] As an improvement of the present invention, the syntactic analysis adopts a component syntactic analysis method to perform syntactic analysis on English sentences, starting from the words of the sentence and gradually merging upwards, building a larger component structure according to the grammatical rules and language knowledge of metaphor until a complete sentence structure is formed, and finally discovering potential metaphors by analyzing the combination of words.

[0008] As an improvement of the present invention, the semantic understanding adopts a combination of the WordNet vocabulary library and the lexical semantic similarity algorithm to understand the semantics, uses the synonym set and the hyponym relationship information in WordNet to analyze the semantics of the words, and calculates the semantic similarity between the words corresponding to the subject and the object in the metaphor by means of semantic similarity calculation. If the similarity is low, but there is a certain connection in a specific context, it may be a metaphorical relationship.

[0009] As an improvement of the present invention, the context is divided into short-term context and long-term context, the metaphor rules are divided into implicit rules and similarity rules, the metaphor rule set is generated according to the metaphor rules, and the lexical, grammatical and semantic analysis of English sentences is combined through machine learning training based on large data sets to extract the literal meaning of rhetorical expressions in English sentences, and then find out the location points where the literal meaning is blocked.

[0010] The present invention also provides a system for quickly extracting metaphors in an English assessment process based on a large language model, comprising: The data preprocessing module is used to clean, segment, and mark the parts of speech of the input text, audio, and video, remove noise data and irregular expressions, convert the input content into a format suitable for model processing, and establish a vocabulary to map each word to a unique integer number so that the model can perform numerical calculations; The feature extraction module is used to process the English words in the input text sequence one by one through the LSTM technology in the context information extraction, and update the current hidden state by combining the current input and the hidden state at the previous moment; The data training module is used to train the model using large-scale English data. Through the back-propagation algorithm, the model parameters are continuously adjusted to minimize the loss function and improve the data training effect. The reasoning parsing module is used to encode the input content through the encoder and extract feature information. Then, the decoder gradually generates parsing results based on these feature information and uses the Viterbi algorithm to find the most likely state sequence. Starting from the starting state, the probability of each state is gradually calculated based on the observation and state transition probability and observation probability, and the state sequence with the largest probability is selected as the parsing result. The output metaphor dataset module is used to convert the input content into a metaphor dataset according to the grammatical rules of metaphor and store it; Preferably, the system for quickly extracting metaphors in the English assessment process based on a large language model also includes a scoring module for scoring metaphor results. The scoring criteria are divided into four types: vocabulary familiarity, vocabulary polysemy, sentence structure complexity, and semantic rule deviation. Each type is 0-10 points, and the final score is a summary method.

[0011] The present invention also provides a computing device, comprising a memory and one or more processors, wherein the memory stores executable code, and when the one or more processors execute the executable code, they are used to implement the method for quickly extracting metaphors in the English assessment process based on a large language model as described in any one of claims 1 to 6.

[0012] The present invention also provides a computer product storing a computer program. When the computer program is executed by a processor, the method for quickly extracting metaphors in an English assessment process based on a large language model as described in any one of claims 1 to 6 is implemented.

[0013] In addition to the traditional metaphor extraction method in the English assessment process, the present invention provides a new extraction and evaluation method to quickly extract metaphors automatically, reduce labor input costs, and increase evaluation efficiency. The technical solution provided by the present invention can have the following beneficial effects: (1) The present invention can automatically extract metaphorical expressions from a large amount of English text through a deep learning algorithm of a large language model. Without human intervention, the system can quickly process and analyze input data and identify metaphors therein. Support multimodal input: the system can process not only plain text data, but also voice and video data, converting them into text for metaphor recognition. This greatly expands the application scope of the system, making it suitable for various forms of English assessment requests.

[0014] (2) The system provided by the present invention can combine short-term and long-term context information to more accurately identify metaphors. Short-term context refers to the grammatical structure and lexical relations within a sentence, while long-term context involves the semantic understanding of a paragraph or the entire text. This dual context-awareness capability improves the accuracy of metaphor recognition. Through lexical analysis, syntactic analysis, and semantic understanding, the system can deeply mine metaphorical expressions in the text to ensure accurate recognition of metaphors in a variety of complex contexts.

[0015] (3) The system of the present invention has a rich built-in metaphor rule library, including implicit rules and similarity rules. Implicit rules convey meaning through suggestion and association, while similarity rules make metaphorical expressions based on the similarities between two things. Based on these rules, the system can intelligently match and infer metaphorical expressions in the text and output metaphorical results. This not only improves the accuracy of metaphor recognition, but also enables the system to adapt to different types of metaphorical expressions.

[0016] (4) The system provided by the present invention supports input data in multiple formats such as text, voice and video. For voice and video data, the system first converts them into text and then performs metaphor recognition. This multimodal data processing capability enables the system to be widely used in various English assessment scenarios. Through steps such as word segmentation, part of speech tagging, syntactic analysis and semantic understanding, the system can generate detailed lexical data, syntactic data and semantic understanding data, providing a solid foundation for feature extraction of large language models.

[0017] (5) When metaphor recognition fails, the system of the present invention can automatically complete and optimize the content of the English assessment request through the reasoning and parsing module of the large language model. The system will generate modification suggestions based on the metaphor rules and automatically generate text that conforms to the metaphor rules. This automatic completion and optimization function not only improves the accuracy of metaphor recognition, but also provides users with a more intelligent English assessment experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0019] Figure 1 A flowchart of a method for quickly extracting metaphors in an English assessment process based on a large language model provided by an embodiment of the present invention.

[0020] Figure 2 A schematic diagram of the modules of a system for quickly extracting metaphors in an English assessment process based on a large language model provided by an embodiment of the present invention.

[0021] Figure 3 A schematic diagram of a method for quickly extracting metaphors in an English assessment process based on a large language model provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0022] The present invention will be further explained below in conjunction with specific implementation modes. It should be understood that the following specific implementation modes are only used to illustrate the present invention and are not used to limit the scope of the present invention.

[0023] The terms used in this specification are only for the purpose of describing specific embodiments and are not intended to limit this specification. The singular forms "a", "the" and "the" used in this specification are also intended to include plural forms unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used in this article refers to and includes any or all possible combinations of one or more associated listed items.

[0024] It should be understood that although the terms first, second, third, etc. may be used in this specification to describe various information, this information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of this specification, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0025] In addition, the described features, structures or characteristics may be combined in one or more embodiments in any suitable manner. In the following description, many specific details are provided to provide a full understanding of the embodiments of this specification sheet. However, those skilled in the art will appreciate that the technical solutions of this specification sheet may be put into practice without one or more of the specific details, or other methods, components, devices, steps, etc. may be adopted. In other cases, known methods, devices, implementations or operations are not shown or described in detail to avoid blurring the various aspects of this specification sheet.

[0026] The method for quickly extracting metaphors in an English assessment process based on a large language model provided in an embodiment of the present invention aims to improve the efficiency and accuracy of metaphor recognition and provide more precise technical support for English teaching and language comprehension. Specifically, the method includes the following steps: S1. After receiving the English evaluation request for metaphor extraction, the grammar processing module processes the evaluation request. The content of the evaluation request is multimodal, which can be text, voice, or video. After processing by the grammar processing module, lexical data, syntactic data, and semantic understanding data are formed. Lexical data is obtained through lexical analysis to form word segmentation results and part-of-speech tagging results. Syntactic data is obtained through syntactic analysis to obtain grammatical structure recognition results and short sentence recognition results. Semantic understanding data is obtained through semantic understanding to obtain lexical semantic parsing results and sentence semantic understanding results.

[0027] If the processing is successful, enter S2, otherwise, throw an exception.

[0028] The grammar processing module constructed in the embodiment has three processing modes, namely lexical analysis, syntactic analysis and semantic understanding. Among them, the lexical analysis performs word segmentation and part-of-speech tagging on the content of the English assessment request and outputs the result, wherein the word segmentation splits the input text into individual words or phrases, for example, "The cat is sleeping on the mat." will be split into "The", "cat", "is", "sleeping", "on", "the", "mat". Part-of-speech tagging marks the part of speech of each word after word segmentation, such as noun, verb, adjective, adverb, etc., where "cat" is a noun and "sleeping" is a verb. The implementation process of the lexical analysis related technology is as follows: a rule-based word segmentation algorithm and a statistical-based word segmentation algorithm are adopted, wherein the rule-based word segmentation algorithm adopts the forward maximum matching method to match English words, and takes a number of characters in the English corpus to be segmented from left to right to form a candidate word substring. If the substring matches a word in the vocabulary, it is segmented as a word. This process is repeated until the entire corpus is segmented; the statistical-based word segmentation algorithm adopts the N-gram language model, which performs word segmentation by counting the frequency of N consecutive words in the English corpus and counting the probability of multiple consecutive words appearing, thereby determining the best word segmentation position.

[0029] Syntactic analysis recognizes the grammatical structure and phrases of the English assessment request content, analyzes the grammatical structure of the sentence, and determines the subject, predicate, object and other components. For example, in "The cat is sleeping on the mat.", "The cat" is the subject, "is sleeping" is the predicate, and "on the mat" is the adverbial. It also recognizes various phrase structures, such as noun phrases and verb phrases. For example, "the sleeping cat" is a noun phrase. The relevant technologies of syntactic analysis are implemented as follows: English sentences are analyzed using the component syntactic analysis method, starting from the words of the sentence, gradually merging upwards, and building a larger component structure according to the grammatical rules and language knowledge of metaphors until a complete sentence structure is formed. Finally, potential metaphors are discovered by analyzing the combination of words.

[0030] Semantic understanding performs lexical semantic analysis and sentence semantic understanding on the content of the English assessment request, understands the specific meaning of each word and sentence, and infers the meaning of the words and sentences by analyzing the relationship between words and grammatical structure. The relevant technologies for semantic understanding are as follows: semantics are understood by combining the WordNet vocabulary library and the lexical semantic similarity algorithm, and the synonym set, hyponym relationship and other information in WordNet are used to analyze the semantics of words. The semantic similarity between the words corresponding to the subject and the object in the metaphor is calculated by semantic similarity calculation. If the similarity is low, but there is a certain connection in a specific context, it may be a metaphorical relationship.

[0031] S2. After successful processing, the lexical data, syntactic data and semantic understanding data enter the large language model for feature extraction. The feature extraction module analyzes and processes the above data to form context information. The context information includes short-term context information and long-term context information. By matching the context information and metaphor rules, the metaphorical expressions in the English assessment request are identified. According to the metaphor rules, whether there are metaphors between sentences is identified, and the metaphor results are output.

[0032] If the metaphor result is output successfully, go to S3; otherwise, throw an exception.

[0033] The context constructed by the embodiment is divided into short-term context and long-term context. The short-term context considers several sentences before and after the current sentence to better understand the meaning of the current sentence. The long-term context ensures the content coherence during the parsing process based on the information between sentences.

[0034] In step S2, the metaphor rule is divided into implicit rules and similarity rules. Implicit rules do not use obvious metaphors such as "like" and "such as", but convey meaning through suggestion and association, such as "Time is a thief." Here, time is directly called a thief without using words such as "like" to connect. Similarity rules are based on the similarity between two different things. This similarity can be in terms of appearance, nature, function, etc., such as "Her eyes are jewels." Her eyes are gems, and eyes and gems are similar in terms of brightness and preciousness.

[0035] S3. After the metaphor result is output, it is matched with the metaphor rule set. The metaphor rule set comes from the large language model's learning of a large amount of English corpus. The specific feature is that metaphor describes one thing as another thing. Common metaphors include extended metaphor, metonymy, mixed metaphor, absolute metaphor, allusion, idiom, allusion, metaphor, source metaphor, new metaphor, latent metaphor, old metaphor, conceptual metaphor, extreme metaphor, simple metaphor, compound metaphor, complex metaphor, etc.

[0036] If the metaphor matches successfully, go to S4, otherwise, go to S5.

[0037] The large language model constructed in the embodiment is divided into four modules, namely, a data preprocessing module, a data training module, a reasoning and parsing module, and an output metaphor data set module.

[0038] The data preprocessing module cleans, segments, and marks the parts of speech of the input text, audio, and video, removes noise data and irregular expressions, converts the input content into a format suitable for model processing, and establishes a vocabulary, mapping each word to a unique integer number to facilitate numerical calculations by the model.

[0039] The data training module uses large-scale English data to train the model. Through the back-propagation algorithm, the model parameters are continuously adjusted to minimize the loss function and improve the data training effect.

[0040] The reasoning parsing module encodes the input content through the encoder to extract feature information. Then, the decoder gradually generates parsing results based on these feature information, such as part-of-speech tagging, syntactic analysis, semantic understanding, etc. The Viterbi algorithm is used to find the most likely state sequence. Starting from the starting state, the probability of each state is gradually calculated based on the observation and state transition probability and observation probability. The state sequence with the highest probability is selected as the parsing result. The final state sequence can correspond to the results of part-of-speech tagging, grammatical analysis, etc.

[0041] The output metaphor dataset module converts the input content into a metaphor dataset according to the grammatical rules of metaphor and stores it.

[0042] S4: After the metaphor is successfully matched, the metaphor evaluation results are manually reviewed and scored, and then enter S6.

[0043] The scoring results constructed in the embodiment are based on the difficulty of metaphor classification, such as extended metaphor, metonymy, mixed metaphor, absolute metaphor, metaphor, idiom, and implicit metaphor, which are scored as five points. Synecdoche, source metaphor, new metaphor, latent metaphor, and old metaphor are scored as four points. Conceptual metaphor, extreme metaphor, simple metaphor, compound metaphor, and complex metaphor are scored as three points.

[0044] S5: If the evaluation fails, metaphors are automatically generated for the wrong output through the metaphor dataset. First, the metaphor dataset is compared with the result after feature extraction, and the metaphor rules are matched to generate lexical and syntactic structures that do not meet the conditions. The metaphor dataset generates modification suggestions based on the metaphor rules. These suggestions may include replacing vocabulary, changing grammatical structures, introducing metaphors, etc. After executing the suggestions, a sentence that compiles the original sentence meaning and contains metaphors is generated, and then enters S6.

[0045] The modification suggestions constructed by the embodiment are automatically generated metaphors for erroneous outputs through metaphor datasets.

[0046] For example, the original sentence that failed the assessment was “He is very angry.” He is very angry.

[0047] The metaphor dataset generates modification suggestions based on metaphor rules. The modification suggestions are: "He is a boilingvolcano." He is a boilingvolcano. Further optimization of the metaphor: It can be adjusted to "He is like a boilingvolcano, ready to erupt at any moment." He is like a boilingvolcano, ready to erupt at any moment. This more vividly expresses his angry state, and the automatic generation is completed.

[0048] S6: The process ends.

[0049] Through the above steps, metaphors can be quickly and automatically extracted, reducing labor input costs and increasing evaluation efficiency.

[0050] The embodiment of the present invention also provides a system for quickly extracting metaphors in the English assessment process based on a large language model. The large language model of the system is divided into four modules, namely, a data preprocessing module, a data training module, a reasoning and parsing module, and an output metaphor data set module. Among them: The data preprocessing module cleans, segments, and marks the parts of speech of the input text, audio, and video to remove noise data and irregular expressions, convert the input content into a format suitable for model processing, and establish a vocabulary to map each word to a unique integer number so that the model can perform numerical calculations.

[0051] The data training module adopts the deep neural network technology in the Transformer architecture and uses large-scale English data to train the model. Through the back-propagation algorithm, the model parameters are continuously adjusted, including the learning rate, batch size and number of training rounds. The learning rate is set to 0.01-0.001, the batch size is set to 32-64, and the number of training rounds is 200-500. Finally, the result of minimizing the loss function is determined to obtain the appropriate training effect.

[0052] The reasoning parsing module encodes the input content through the encoder to extract feature information. The encoder is based on the Transformer structure and consists of multiple stacked encoding layers. Each encoding layer contains a multi-head self-attention mechanism, which can focus on different parts of words, sentences, and sentences at the same time, effectively capture long-distance dependencies, and thus more comprehensively extract the semantic features and structural features of multimodal English corpus. In addition, the encoding layer is also equipped with a feedforward neural network, which further enhances the model's feature extraction ability for multimodal English corpus through nonlinear transformation. After being processed by the encoder, the input content is converted into a high-dimensional feature vector representation. The Viterbi algorithm is used to find the most likely state sequence. In the English part-of-speech tagging task, the hidden state can be different parts of speech, and the observed state can be a word. Through the Viterbi algorithm, the most likely part-of-speech sequence can be found based on a given word sequence, thereby achieving accurate part-of-speech tagging. Starting from the starting state, the probability of each state is gradually calculated based on the observation and state transition probability and the observation probability, and the state sequence with the largest probability is selected as the parsing result. The final state sequence can correspond to the results of part-of-speech tagging, grammatical analysis, etc.

[0053] The output metaphor dataset module converts the input content into a metaphor dataset according to the grammatical rules of metaphor and stores it. The technical implementation of the metaphor dataset is as follows: the metaphor rule set is generated according to the metaphor rule. Metaphor is manifested as a grammatical transfer rule in linguistics. The formula of the metaphor rule is F (Figurative) = "like L (Literal)", where F represents the rhetorical meaning of English words, L represents the literal meaning of the English word itself, and like represents the derivation of the literal meaning in the English sentence. Through machine learning training based on large data sets, combined with the analysis of the lexical, grammatical and semantic aspects of English sentences, the literal meaning of the rhetorical expressions in the English sentences is extracted, and then the location points where the literal meaning is blocked are found, such as Figure 3 As shown in the figure, “lovers” can really bathe, just as they can really experience pleasure. Therefore, “But ye lovers, that...” is just a context introduced for “bathe in gladnesse” - the “ditto symbol” strictly indicates that it is not part of the metaphor. On the other hand, the “blank part” (shown by the horizontal line) symbolizes the textual gaps of literal or rhetorical interpretation. After the extraction is completed, the metaphor data is stored in the database to form a metaphor dataset.

[0054] The system described in the embodiment of the present invention can improve the efficiency and accuracy of metaphor extraction. The large language model can quickly process a large amount of text and accurately identify metaphorical expressions. It provides more comprehensive indicators for English assessment. By extracting and analyzing metaphors, students' language comprehension and application abilities can be more deeply understood. It is helpful for English teaching. Teachers can carry out targeted teaching based on the metaphor extraction report to improve students' understanding and application abilities of metaphors.

[0055] A computing device includes a memory and one or more processors, wherein the memory stores executable code, and is characterized in that when the one or more processors execute the executable code, they are used to implement a method for quickly extracting metaphors in an English assessment process based on a large language model.

[0056] The computing device provided in the embodiment, in addition to the processor and memory, also includes hardware required for other services such as internal bus, network interface, memory, etc. at the hardware level. The memory is a non-volatile memory, and the processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to implement the evaluation method of the implicit semantic recognition ability of the large language model described in S1-S7 above. Of course, in addition to the software implementation method, the present invention does not exclude other implementation methods, such as logic devices or a combination of software and hardware, etc., that is to say, the execution subject of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0057] Based on the same inventive concept, an embodiment further provides a computer product storing a computer program, which, when executed by a processor, implements a method for quickly extracting metaphors in an English assessment process based on a large language model.

[0058] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that several improvements and modifications may be made without departing from the principles of the present invention, and these improvements and modifications all fall within the scope of protection of the claims of the present invention.

Claims

1. A method for quickly extracting metaphors in English assessment based on a large language model, characterized in that: The following steps are involved: S1. Process the evaluation request to form lexical data, syntactic data and semantic understanding data; S2. Extract features from the lexical data, syntactic data, and semantic understanding data processed by S1 to form context information. By matching the context information with metaphor rules, identify metaphorical expressions in the English assessment request. According to the metaphor rules, identify whether there are metaphors between sentences and output metaphor results. S3. Match the metaphor result output by S2 with the metaphor rule set. If the metaphor match succeeds, manually review the metaphor evaluation result and give a score. If the metaphor match fails, proceed to S4. S4. Metaphors are automatically generated for incorrect outputs through metaphor datasets. First, the resulting metaphor dataset after feature extraction is compared to match metaphor rules and generate lexical and syntactic sentences that do not meet the conditions. The metaphor dataset generates modification suggestions based on metaphor rules. After executing the suggestions, a sentence that conforms to the meaning of the original sentence and contains metaphors is generated.

2. The method for quickly extracting metaphors in English assessment based on a large language model according to claim 1 is characterized in that: Step S1 specifically includes the following contents: The evaluation request content is multimodal and can be text, voice, or video; After the evaluation request is processed, the lexical data is obtained through lexical analysis to form word segmentation results and part-of-speech tagging results. The syntactic data is obtained through syntactic analysis to obtain grammatical structure recognition results and short sentence recognition results. The semantic understanding data is obtained through semantic understanding, and a lexical semantic analysis result and a sentence semantic understanding result are obtained.

3. The method for quickly extracting metaphors in English assessment based on a large language model according to claim 2 is characterized in that: The lexical analysis adopts a rule-based word segmentation algorithm and a statistics-based word segmentation algorithm, wherein: The rule-based word segmentation algorithm adopts the forward maximum matching method to match English words, taking a number of characters in the English corpus to be segmented from left to right to form a candidate word substring. If the substring matches a word in the vocabulary, it is segmented as a word, and this process is repeated until the entire corpus is segmented; The statistical-based word segmentation algorithm adopts the N-gram language model, performs word segmentation by counting the frequencies of N consecutive words in an English corpus, and counts the probabilities of multiple consecutive words appearing, thereby determining the best word segmentation position.

4. The method for quickly extracting metaphors in English assessment based on a large language model according to claim 2 is characterized in that: The syntactic analysis adopts the component syntactic analysis method to perform syntactic analysis on English sentences, starting from the words of the sentence and gradually merging upwards, building a larger component structure according to the grammatical rules and language knowledge of metaphor until a complete sentence structure is formed, and finally discovering potential metaphors by analyzing the combination of words.

5. The method for quickly extracting metaphors in English assessment based on a large language model according to claim 2 is characterized in that: The semantic understanding adopts a combination of the WordNet vocabulary database and the lexical semantic similarity algorithm to understand the semantics, uses the synonym set and the hyponym relationship information in WordNet to analyze the semantics of words, and calculates the semantic similarity between the words corresponding to the subject and the object in the metaphor by means of semantic similarity calculation. If the similarity is low, but there is a certain connection in a specific context, it may be a metaphorical relationship.

6. The method for quickly extracting metaphors in English assessment based on a large language model according to claim 1, characterized in that: The context is divided into short-term context and long-term context, the metaphor rules are divided into implicit rules and similarity rules, and the metaphor rule set is generated according to the metaphor rules. Through machine learning training based on large data sets, the lexical, grammatical and semantic analysis of English sentences is combined to extract the literal meaning of rhetorical expressions in English sentences, and then find out the location points where the literal meaning is blocked.

7. A system for quickly extracting metaphors in English assessment based on a large language model, characterized in that: include: The data preprocessing module is used to clean, segment, and mark the parts of speech of the input text, audio, and video, remove noise data and irregular expressions, convert the input content into a format suitable for model processing, and establish a vocabulary to map each word to a unique integer number so that the model can perform numerical calculations; The feature extraction module is used to process the English words in the input text sequence one by one through the LSTM technology in the context information extraction, and update the current hidden state by combining the current input and the hidden state at the previous moment; The data training module is used to train the model using large-scale English data. Through the back-propagation algorithm, the model parameters are continuously adjusted to minimize the loss function and improve the data training effect. The reasoning parsing module is used to encode the input content through the encoder and extract feature information. Then, the decoder gradually generates parsing results based on these feature information and uses the Viterbi algorithm to find the most likely state sequence. Starting from the starting state, the probability of each state is gradually calculated based on the observation and state transition probability and observation probability, and the state sequence with the largest probability is selected as the parsing result. The output metaphor dataset module is used to convert the input content into a metaphor dataset according to the grammatical rules of metaphor and store it.

8. A computing device comprising a memory and one or more processors, wherein the memory stores executable code, characterized in that: When the one or more processors execute the executable code, they are used to implement the method for quickly extracting metaphors in an English assessment process based on a large language model as described in any one of claims 1 to 6.

9. A computer product storing a computer program, characterized in that: When the computer program is executed by a processor, the method for quickly extracting metaphors in an English assessment process based on a large language model as described in any one of claims 1 to 6 is implemented.

Images