Foreign language teaching effect evaluation method and system based on AI

Abstract

The invention relates to the technical field of teaching effect evaluation, in particular to an AI-based foreign language teaching effect evaluation method and system. The method comprises the following steps: performing text paragraph logic segmentation on a foreign language writing examination content text to obtain paragraph logic segmentation data of the foreign language writing examination content; performing native language system interference identification on the foreign language writing assessment content text based on the paragraph logic segmentation data to obtain content paragraph native language system interference identification data; performing adaptive language system anomaly recognition optimization according to the content paragraph native language system interference recognition data to obtain an AI language system anomaly recognition architecture; and performing foreign language teaching effect evaluation according to the AI language system abnormity identification architecture to obtain foreign language teaching effect evaluation data, and sending the foreign language teaching effect evaluation data to the terminal. The AI foreign language teaching effect evaluation method is optimized, so that the AI foreign language teaching effect evaluation method is more perfect.

Description

Technical Field

[0001] The present invention relates to the technical field of teaching effect evaluation, and in particular to an AI-based foreign language teaching effect evaluation method and system. Background Art

[0002] Previous foreign language teaching evaluation methods mostly relied on teachers' subjective judgment and students' language ability tests. However, these methods often fail to fully and accurately reflect students' real language level and the difficulties they face in the process of foreign language learning. With the continuous advancement of artificial intelligence (AI) technology, especially the development of natural language processing (NLP) and machine learning, AI-based foreign language teaching effect evaluation methods have emerged. This method can analyze students' language output, identify their language errors, interference factors, and learning progress, and thus provide a more objective and accurate teaching evaluation. However, a traditional AI-based foreign language teaching effect evaluation method has the problem of inaccurate logical analysis of students' writing information and large errors in identifying the influence of students' native language structure, resulting in low accuracy in teaching effect evaluation. Summary of the invention

[0003] Based on this, it is necessary to provide an AI-based foreign language teaching effectiveness evaluation method to solve at least one of the above technical problems.

[0004] To achieve the above purpose, a foreign language teaching effect evaluation method based on AI is provided, the method comprising the following steps: Step S1: obtaining the foreign language writing assessment content text submitted by the student; performing logical segmentation of the foreign language writing assessment content text into paragraphs to obtain paragraph logical segmentation data of the foreign language writing assessment content; Step S2: Based on the paragraph logic segmentation data, the information entropy sequence interference gradient quantification is performed on the foreign language writing assessment content text to obtain the information entropy sequence interference gradient data; based on the information entropy sequence interference gradient data, the native language system interference recognition is performed on the foreign language writing assessment content text to obtain the content paragraph native language system interference recognition data; Step S3: Obtain the underlying learning architecture of the AI ​​tool; perform adaptive language system anomaly recognition optimization on the underlying learning architecture of the AI ​​tool according to the native language system interference recognition data of the content paragraphs, and obtain the AI ​​language system anomaly recognition architecture; Step S4: Perform foreign language teaching effectiveness evaluation on the foreign language writing assessment content text according to the AI ​​language system anomaly recognition architecture, obtain foreign language teaching effectiveness evaluation data, and send the foreign language teaching effectiveness evaluation data to the terminal.

[0005] Preferably, step S1 comprises the following steps: Step S11: Obtaining the foreign language writing test content text submitted by the student; Step S12: cleaning the foreign language writing assessment content text to obtain a cleaned foreign language writing assessment content text; Step S13: Performing a chapter structure analysis on the cleaned text of the foreign language writing assessment content to obtain the chapter structure data of the writing content; Step S14: logically segment the cleaned text of the foreign language writing assessment content according to the chapter structure narrative data of the writing content to obtain logical segmentation data of the foreign language writing assessment content.

[0006] Preferably, step S2 comprises the following steps: Step S21: performing sentence parallel / subordinate connection analysis on the foreign language writing assessment content text according to the paragraph logic segmentation data to obtain sentence parallel / subordinate connection data; Step S22: Based on the sentence parallel / subordinate connection data and the paragraph logic segmentation data, the foreign language writing test content text is identified with logic jump links to obtain content paragraph logic jump data; Step S23: quantifying the information entropy sequence interference gradient of the foreign language writing test content text based on the content paragraph logic jump data to obtain information entropy sequence interference gradient data; Step S24: performing native language system interference recognition on the foreign language writing test content text according to the information entropy sequence interference gradient data and the content paragraph logical jump data to obtain content paragraph native language system interference recognition data.

[0007] Preferably, step S23 includes the following steps: Step S231: Count the frequency of logical jump connectives in the foreign language writing assessment content text based on the logical jump data of the content paragraphs to obtain the frequency data of logical jump connectives; Step S232: Decomposing the semantic span word vector of the connective word through the frequency data of the logical jump connective word to obtain the semantic span word vector of the connective word; Step S233: performing a grammatical fuzziness regression analysis of the previous and next paragraphs on the foreign language writing assessment content text according to the semantic span word vector of the connective and the frequency data of the logical jump connective to obtain the grammatical fuzziness regression data of the previous and next paragraphs; Step S234: performing a simulation analysis of the disordered state of the dimensionality constraint of the information entropy sequence based on the semantic span word vector of the connective and the fuzzy regression data of the preceding and following paragraphs, and obtaining the disordered state of the dimensionality constraint of the information entropy sequence; Step S235: quantizing the information entropy sequence interference gradient of the information entropy sequence dimension constraint disorder state to obtain the information entropy sequence interference gradient data.

[0008] Preferably, step S234 includes the following steps: Perform polysemous clustering analysis on the semantic span word vectors of connectives to obtain polysemous clustering data of connectives; Perform grammatical boundary fuzzy trend analysis on the grammatical fuzzy regression data of the previous and next paragraphs to obtain the grammatical boundary fuzzy trend data of the previous and next paragraphs; Based on the polysemous clustering data of connectives and the fuzzy trend data of the grammatical boundaries of the previous and next paragraphs, the context change boundary fitting is performed to obtain the context change boundary fitting data of the previous and next paragraphs; Based on the Markov chain, the context boundary fitting state is simulated for the context change boundary fitting data before and after the paragraph to obtain the context boundary state fitting data; According to the context boundary state fitting data, the dimension constraint disorder state of the information entropy sequence is simulated and analyzed to obtain the dimension constraint disorder state of the information entropy sequence.

[0009] Preferably, step S24 includes the following steps: Step S241: Acquire native language architecture data of different students; Step S242: performing native language migration feature analysis on the foreign language writing assessment content text according to the content paragraph logic jump data and the native language system structure data to obtain the assessment content native language migration feature data; Step S243: Identify syntactic structure logic deviation based on the native language transfer feature data of the test content and the content paragraph logic jump data to obtain native language syntactic structure logic deviation data; Step S244: performing semantic deviation recognition on the native language transfer feature data of the assessment content to obtain native language semantic feature deviation data; Step S245: mapping the foreign language writing assessment content text to a native language dominant syntactic framework according to the native language syntactic structure logic deviation data and the native language ideographic feature deviation data to obtain native language dominant syntactic framework mapping data; Step S246: Perform native language system interference identification based on the information entropy sequence interference gradient data and the native language dominant syntactic framework mapping data to obtain content paragraph native language system interference identification data.

[0010] Preferably, step S3 comprises the following steps: Step S31: Obtain the underlying learning architecture of the AI ​​tool; Step S32: normalizing the information entropy sequence interference gradient data to obtain information entropy sequence interference gradient normalized data; Step S33: Perform AI architecture disorder degree recognition learning reinforcement on the underlying learning architecture of the AI ​​tool according to the information entropy sequence interference gradient normalization data to obtain an AI disorder degree recognition learning reinforcement architecture; Step S34: Based on the content paragraph native language system interference recognition data, the AI ​​disorder degree recognition learning reinforcement architecture is optimized for adaptive language system anomaly recognition to obtain the AI ​​language system anomaly recognition architecture.

[0011] Preferably, step S33 includes the following steps: Step S331: performing topological information entropy discretization processing on the information entropy sequence interference gradient normalized data to obtain an information entropy topological discrete feature space; Step S332: performing feature orthogonal mapping and information entropy reconstruction processing on the information entropy topological discrete feature space to obtain a feature mapping information entropy tensor; Step S333: performing interference gradient calibration processing according to the feature mapping information entropy tensor to obtain interference gradient calibration data; Step S334: Based on the interference gradient calibration data, the underlying learning architecture of the AI ​​tool is subjected to AI architecture disorder degree recognition learning reinforcement to obtain an AI disorder degree recognition learning reinforcement architecture.

[0012] Preferably, step S34 includes the following steps: Step S341: performing native language part-of-speech interference anomaly logic analysis on the native language system interference recognition data of the content paragraph to obtain native language part-of-speech interference anomaly logic data; Step S342: Summarizing the part-of-speech logic metaphor deviation based on the native language part-of-speech interference abnormal logic data to obtain the part-of-speech logic metaphor deviation data; Step S343: performing concept cross-domain metaphor anomaly analysis on the part-of-speech logic metaphor deviation data to obtain concept cross-domain metaphor anomaly data; Step S344: Based on the part-of-speech logic metaphor deviation data and the concept cross-domain metaphor anomaly data, the AI ​​disorder degree recognition learning reinforcement architecture is adaptively optimized for language system anomaly recognition to obtain the AI ​​language system anomaly recognition architecture.

[0013] Preferably, the present invention further provides an AI-based foreign language teaching effect evaluation system, which is used to execute the AI-based foreign language teaching effect evaluation method as described above. The AI-based foreign language teaching effect evaluation system includes: The paragraph logic segmentation module is used to obtain the foreign language writing assessment content text submitted by the students; perform paragraph logic segmentation on the foreign language writing assessment content text to obtain paragraph logic segmentation data of the foreign language writing assessment content; The interference recognition module is used to quantify the information entropy sequence interference gradient of the foreign language writing assessment content text based on the paragraph logic segmentation data to obtain the information entropy sequence interference gradient data; perform native language system interference recognition on the foreign language writing assessment content text based on the information entropy sequence interference gradient data to obtain the content paragraph native language system interference recognition data; The anomaly recognition optimization module is used to obtain the underlying learning architecture of the AI ​​tool; the underlying learning architecture of the AI ​​tool is optimized for adaptive language system anomaly recognition based on the native language system interference recognition data of the content paragraphs to obtain the AI ​​language system anomaly recognition architecture; The effect evaluation module is used to evaluate the foreign language teaching effect on the foreign language writing assessment content text according to the AI ​​language system anomaly recognition architecture, obtain the foreign language teaching effect evaluation data, and send the foreign language teaching effect evaluation data to the terminal.

[0014] The beneficial effect of the present invention is that, first, by obtaining the foreign language writing assessment content text submitted by the student, the logical segmentation of the text paragraph is performed. The key to this process is to decompose the student's writing content into a plurality of logically clear paragraphs as a whole, so that each paragraph can independently express a specific idea or theme. Through this segmentation, the structure of the text becomes clearer, which helps to more accurately identify the language characteristics, logical organization and expression problems of different paragraphs in subsequent analysis, and provides accurate basic data for subsequent interference identification and evaluation. Based on the paragraph logic segmentation data obtained in step S1, the information entropy sequence interference gradient quantification process quantifies the language characteristics of each paragraph by analyzing the information density and complexity in the text. Through the change of entropy value, it can be revealed which parts of the text have logical jumps or unnatural language use. In addition, according to the information entropy sequence interference gradient data, the system can identify foreign language expression problems caused by native language interference. This recognition helps detect the native language traces that appear in students' writing, and then provides a basis for subsequent teaching intervention, effectively improving the pertinence of teaching. The underlying learning architecture of the AI ​​tool is adaptively optimized by analyzing the native language system interference recognition data. The AI ​​system adjusts the language system anomaly recognition architecture according to different student writing characteristics and native language interference patterns. By continuously learning and optimizing its recognition capabilities, AI can improve its sensitivity to subtle problems in foreign language writing, especially for language habits and improper structures, and can make more accurate judgments. This process provides the AI ​​system with a dynamic adjustment capability to ensure efficient and accurate language anomaly recognition in different student learning scenarios. Finally, the optimized AI language system anomaly recognition architecture will evaluate the teaching effect of students' foreign language writing assessment content. This step comprehensively analyzes the language characteristics, logical structure and native language interference in students' writing, so that the system can objectively and comprehensively evaluate the effectiveness of foreign language teaching. The evaluation results will specifically reflect the obstacles or misunderstandings in the students' language expression, and put forward targeted improvement suggestions. The evaluation data is fed back to teachers and students through the terminal to help teachers adjust their teaching strategies, and students can improve their language skills in a targeted manner based on the evaluation results, thereby effectively improving the effectiveness of foreign language learning. Therefore, the present invention optimizes a traditional AI-based foreign language teaching effect evaluation method, solves the problems of inaccurate logical analysis of students' writing information and large errors in identifying the influence of students' native language structure, resulting in low accuracy in teaching effect evaluation, improves the accuracy of logical analysis of students' writing information, reduces the errors in identifying the influence of students' native language structure, and improves the accuracy of teaching effect evaluation. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1A flowchart of a method for evaluating the effectiveness of foreign language teaching based on AI; Figure 2 for Figure 1 Detailed implementation steps of step S2 in the flowchart; Figure 3 for Figure 1 Detailed implementation steps of step S3 in FIG. DETAILED DESCRIPTION

[0016] See also Figures 1 to 3 , a foreign language teaching effect evaluation method based on AI, the method comprising the following steps: Step S1: obtaining the foreign language writing assessment content text submitted by the student; performing logical segmentation of the foreign language writing assessment content text into paragraphs to obtain paragraph logical segmentation data of the foreign language writing assessment content; Step S2: Based on the paragraph logic segmentation data, the information entropy sequence interference gradient quantification is performed on the foreign language writing assessment content text to obtain the information entropy sequence interference gradient data; based on the information entropy sequence interference gradient data, the native language system interference recognition is performed on the foreign language writing assessment content text to obtain the content paragraph native language system interference recognition data; Step S3: Obtain the underlying learning architecture of the AI ​​tool; perform adaptive language system anomaly recognition optimization on the underlying learning architecture of the AI ​​tool according to the native language system interference recognition data of the content paragraphs, and obtain the AI ​​language system anomaly recognition architecture; Step S4: Perform foreign language teaching effectiveness evaluation on the foreign language writing assessment content text according to the AI ​​language system anomaly recognition architecture, obtain foreign language teaching effectiveness evaluation data, and send the foreign language teaching effectiveness evaluation data to the terminal.

[0017] In the embodiment of the present invention, reference Figure 1 The above is a schematic diagram of the steps of an AI-based foreign language teaching effect evaluation method of the present invention. In this example, the AI-based foreign language teaching effect evaluation method includes the following steps: Step S1: obtaining the foreign language writing assessment content text submitted by the student; performing logical segmentation of the foreign language writing assessment content text into paragraphs to obtain paragraph logical segmentation data of the foreign language writing assessment content; In an embodiment of the present invention, first, it is necessary to obtain the foreign language writing assessment content text submitted by the student. This operation can be obtained through a network interface or a file upload method to ensure the integrity and correctness of the text data. Next, the text is logically segmented into paragraphs to ensure that the segmented text paragraphs can reflect the logical relationship of each paragraph in the article. In specific implementation, the text submitted by the student is first subjected to basic text cleaning operations, including the removal of irrelevant characters, the normalization of punctuation marks, and the exclusion of garbled information. Then, an algorithm based on natural language processing is used to segment the text into paragraphs in combination with syntactic analysis and semantic analysis methods. The basis for paragraph segmentation is the logical structure, using technologies such as dependency parsing to identify the topic sentence and core semantics of each paragraph, thereby segmenting paragraphs with independent semantic units. The process of paragraph segmentation requires the use of a hierarchical grammatical rule tree to define the boundaries of the paragraphs based on the logical association between the main clause and the subordinate clause in each paragraph, ensuring that each paragraph can fully convey a set of ideas or arguments. The generation of paragraph segmentation data can be achieved by constructing a text matrix. During this process, the sentence and paragraph structure of the text are converted into the form of a data matrix for subsequent processing.

[0018] Step S2: Based on the paragraph logic segmentation data, the information entropy sequence interference gradient quantification is performed on the foreign language writing assessment content text to obtain the information entropy sequence interference gradient data; based on the information entropy sequence interference gradient data, the native language system interference recognition is performed on the foreign language writing assessment content text to obtain the content paragraph native language system interference recognition data; In the embodiment of the present invention, the information entropy sequence interference gradient is first quantified for the foreign language writing assessment content text based on the paragraph logic segmentation data. To this end, the information entropy calculation method is used to measure the amount of information in each paragraph in the text. Specifically, the word frequency statistics are first performed on each paragraph, the frequency of occurrence of each word is calculated, and the information entropy is calculated using the following formula: ,in, Representing terms The probability of occurrence in a paragraph, is the entropy value of the paragraph, reflecting the complexity and uncertainty of the paragraph. Represents the number of words. Then, based on the calculated information entropy value, the interference gradient of the text is further analyzed. The main purpose of interference gradient quantification is to identify changes in information density in the text, and to obtain interference gradient data of the entropy sequence by calculating the rate of change of the entropy value. The interference gradient represents the degree of distortion of information transmission between paragraphs. Next, native language system interference identification is performed based on the information entropy sequence interference gradient data. By comparing the entropy gradient of the paragraph with the known native language system interference model, the existing native language interference is identified. For example, if the entropy gradient of some paragraphs matches the native language pattern more highly, it indicates that there is native language interference in the paragraph. Through this process, the native language system interference identification data of the content paragraph is finally obtained. This data not only contains the location of the paragraph and the type of interference, but also the intensity and impact of the interference, and is output in the form of structured data.

[0019] Step S3: Obtain the underlying learning architecture of the AI ​​tool; perform adaptive language system anomaly recognition optimization on the underlying learning architecture of the AI ​​tool according to the native language system interference recognition data of the content paragraphs, and obtain the AI ​​language system anomaly recognition architecture; In an embodiment of the present invention, in step S3, the underlying learning architecture of the AI ​​tool is first obtained. Specifically, this architecture can be a neural network model based on deep learning, or a feature extraction method based on traditional machine learning. Assuming that a neural network-based architecture is used, a pre-trained language model (such as BERT, GPT, etc.) must be loaded first. Then, the native language system interference recognition data of the content paragraph obtained in step S2 is used to optimize the adaptive language system anomaly recognition of the underlying learning architecture of the AI ​​tool. In order to achieve this optimization, the model is first fine-tuned according to the interference recognition data, and the parameters of the model are adjusted so that it can better identify language system anomalies in the text. The fine-tuning process can use the gradient descent method, and the specific optimization process is as follows: ,in is the learning weight parameter of the model, is the time step parameter, is the learning rate, is the loss function. In this embodiment, the design of the loss function should be able to effectively capture the characteristics of language system anomalies. During the fine-tuning process, the network weights are mainly updated through the back propagation algorithm, so that the model's ability to recognize native language interference is enhanced. The optimized model can more accurately identify language system anomalies in foreign language writing texts, and ultimately output an optimized AI language system anomaly recognition architecture.

[0020] Step S4: Perform foreign language teaching effectiveness evaluation on the foreign language writing assessment content text according to the AI ​​language system anomaly recognition architecture, obtain foreign language teaching effectiveness evaluation data, and send the foreign language teaching effectiveness evaluation data to the terminal.

[0021] In an embodiment of the present invention, based on the AI ​​language system anomaly recognition architecture, the foreign language writing assessment content text is evaluated for foreign language teaching effectiveness. First, each paragraph is analyzed using the optimized AI language system anomaly recognition architecture to detect language system anomalies in the text. This process mainly evaluates the student's foreign language writing level by calculating the difference between the grammatical, semantic and other features of the paragraph and the standard foreign language model. By analyzing factors such as grammatical errors, inappropriate words, and sentence structures in the text, the foreign language teaching effect evaluation data is finally obtained. The evaluation results will include information such as the type of language errors, error frequency, and severity of errors in each paragraph. These data will be converted into the final teaching effect evaluation score through a standardized scoring model. All calculations and scoring in the evaluation process are completed automatically by algorithms without manual intervention. Finally, the foreign language teaching effect evaluation data is sent to the terminal device through the network for teachers and students to view. The data transmission process uses an encryption algorithm to ensure data security and ensure the privacy and reliability of the evaluation results.

[0022] Step S1 includes the following steps: Step S11: Obtaining the foreign language writing test content text submitted by the student; Step S12: cleaning the foreign language writing assessment content text to obtain a cleaned foreign language writing assessment content text; Step S13: Performing a chapter structure analysis on the cleaned text of the foreign language writing assessment content to obtain the chapter structure data of the writing content; Step S14: logically segment the cleaned text of the foreign language writing assessment content according to the chapter structure narrative data of the writing content to obtain logical segmentation data of the foreign language writing assessment content.

[0023] In the embodiments of the present invention, first, the text of the foreign language writing assessment content submitted by students is obtained from the system. This text is usually in the form of an electronic document, such as standard formats like Word documents, PDF files, or TXT files. By reading the content of the file, it is converted into plain text format. At this time, the text content is extracted through a file parsing tool to ensure format standardization and avoid any potential document encoding issues. After the text is obtained, the character set of the text (such as UTF-8 encoding) is further confirmed to ensure its compatibility and avoid garbled problems during subsequent processing. After obtaining the text, it is stored in a text data structure and prepared to enter the subsequent processing flow. For the processing of different document formats, corresponding parsing libraries are used for adaptation, such as a PDF parsing library or a Word document parsing tool, to ensure that the obtained text data is clear and complete. In step S12, text cleaning of the foreign language writing assessment content text is performed. The goal of text cleaning is to remove irrelevant symbols, formatting errors, and noise data to ensure the accuracy of subsequent analysis. In the specific implementation process, first, all non-language characters in the text are removed, such as page numbers, headers, and footers, and special symbols and punctuation marks are removed using regular expressions. Then, the redundant spaces and line breaks in the text are further removed to ensure that each line of text is valid language data. During the processing, the parts of speech of the words in the text are labeled through a syntax analysis library, and irrelevant stop words, such as "de", "shi", "le", etc., are removed. These words have no practical significance for subsequent analysis. The cleaning process is implemented through text processing tools such as the nltk and re libraries in Python to ensure that only the parts useful for semantic analysis are retained in the finally output text. In addition, during the cleaning process, it is also necessary to ensure the handling of text encoding issues to avoid garbled problems caused by inconsistent character sets. After cleaning, the cleaned text data is generated for subsequent analysis. In step S13, discourse structure narrative analysis is performed. The purpose of discourse structure analysis is to identify the overall organization of the text and the structural levels of each paragraph. Specifically, first, a rule-based discourse analysis algorithm is used to process the cleaned text. This algorithm divides different discourse units based on punctuation marks (such as full stops, commas, question marks, etc.) in the text and the grammatical structure of sentences. During the analysis, dependency syntactic analysis is used to identify the relationships between each sentence in the text. For example, by parsing the basic grammatical components such as the subject, predicate, and object in a sentence, the connectivity between each sentence and the previous and subsequent sentences is further judged. During the discourse structure narrative analysis process, some text-based logical structure models are also combined, such as the common discourse structure of "introduction - body - conclusion", and the discourse structure in the text is inferred through pattern matching technology. This process also depends on the topic words in the text to judge the core topic of each paragraph and its position in the discourse.The final output is a chapter structure narrative data, which records the logical level, theme and syntactic structure of each paragraph, as well as the relationship chain between paragraphs, and provides a reference for the subsequent logical segmentation of text paragraphs. In step S14, the text paragraph logical segmentation is performed based on the chapter structure narrative data of the previous step. In the specific operation, the entire text is first decomposed into multiple paragraphs or chapter units using the aforementioned chapter structure narrative data. The basis for paragraph division is the conversion of the theme in the text, the change of grammatical structure and the logical relationship identified in the chapter structure. For example, if the chapter structure data indicates that a paragraph is a further elaboration or summary of the content of the previous paragraph, it is treated as an independent paragraph. By analyzing the linking words (such as "therefore", "in addition", "for example", etc.) appearing in the text, the logical relationship between sentences is judged, and the boundary of the paragraph is further determined. In order to ensure the accuracy of segmentation, deep learning algorithms such as long short-term memory network (LSTM) or recurrent neural network (RNN) can be used to assist in judging the logical relationship between sentences and paragraphs, and the paragraph boundaries are adjusted according to the prediction results of these algorithms. After segmentation, the starting position, ending position and content of each paragraph will be saved to form paragraph logic segmentation data. This data is usually saved in JSON or CSV format, recording the detailed information of each paragraph, including the content of the paragraph, the position of the paragraph in the text, the theme information of the paragraph, etc. Ultimately, this data provides structured text input for subsequent foreign language teaching effect evaluation.

[0024] Step S2 includes the following steps: Step S21: performing sentence parallel / subordinate connection analysis on the foreign language writing assessment content text according to the paragraph logic segmentation data to obtain sentence parallel / subordinate connection data; Step S22: Based on the sentence parallel / subordinate connection data and the paragraph logic segmentation data, the foreign language writing test content text is identified with logic jump links to obtain content paragraph logic jump data; Step S23: quantifying the information entropy sequence interference gradient of the foreign language writing test content text based on the content paragraph logic jump data to obtain information entropy sequence interference gradient data; Step S24: performing native language system interference recognition on the foreign language writing test content text according to the information entropy sequence interference gradient data and the content paragraph logical jump data to obtain content paragraph native language system interference recognition data.

[0025] As an example of the present invention, refer to Figure 2 As shown, in this example, step S2 includes: Step S21: performing sentence parallel / subordinate connection analysis on the foreign language writing assessment content text according to the paragraph logic segmentation data to obtain sentence parallel / subordinate connection data; In the embodiment of the present invention, in step S21, sentence parallel / subordinate connection analysis is performed, the purpose of which is to identify the sentence structure in the foreign language writing assessment content text, especially the use of parallel and subordinate sentences. First, the syntactic structure of each sentence in the text is parsed by a syntactic analysis tool (such as a tool based on dependency syntactic analysis) to construct a dependency tree in the sentence. In the dependency tree, the relationship between sentence components such as the subject, predicate, and object is indicated. By matching the syntactic analysis results of each sentence, the parallel structure and the subordinate structure are identified. Parallel sentences are usually connected by conjunctions such as "and" and "or" to form an equivalence relationship between sentences; subordinate sentences are connected by subordinate conjunctions such as "because" and "although", and the subordinate clauses are attached to the main sentence to form a hierarchical structure. For the parallel and subordinate relationships of each sentence, annotate and record them in the sentence parallel / subordinate connection data. This data includes the connection type (parallel or subordinate) of each sentence, the position of the conjunction, the connected sentence components, and the grammatical hierarchical relationship between sentences. The tools used may include dependency parsing modules based on natural language processing libraries such as SpaCy to ensure that the analysis results are accurate and comply with grammatical rules.

[0026] Step S22: Based on the sentence parallel / subordinate connection data and the paragraph logic segmentation data, the foreign language writing test content text is identified with logic jump links to obtain content paragraph logic jump data; In the embodiment of the present invention, in step S22, according to the sentence parallel / subordinate connection data and paragraph logic segmentation data obtained in step S21, the foreign language writing assessment content text is subjected to logic jump link identification. First, by analyzing the paragraphs of the text, it is judged whether there is a logic jump between sentences or paragraphs according to the content logic relationship between paragraphs. Logical jumps are usually manifested as a sudden change in the discussion in the text, or a lack of obvious linking words or sentence linking methods between paragraphs. Using a logic link identification algorithm, combined with sentence parallel / subordinate connection data, analyze whether the sentence or paragraph transition between paragraphs is smooth, and check whether there are obvious logical gaps or irrelevant jumps. The algorithm used in this process includes rule-based text relationship reasoning, combining the semantic relationship of the linking words in the sentence and the context to judge the rationality of the logic link. For each paragraph, by matching the subject words and emotional colors in the sentence, it is judged whether there is a jump between the paragraph and the previous and next paragraphs, and then it is determined whether there is a logic jump link. The identified logic jump link will be recorded as content paragraph logic jump data, and the data includes information such as the location where the jump occurs, the paragraph involved, and the type of jump. This data provides a basis for subsequent interference gradient quantification.

[0027] Step S23: quantifying the information entropy sequence interference gradient of the foreign language writing test content text based on the content paragraph logic jump data to obtain information entropy sequence interference gradient data; In the embodiment of the present invention, in step S23, the information entropy sequence interference gradient is quantified according to the content paragraph logic jump data obtained in step S22. Information entropy is a measure of the uncertainty or complexity of information. In this step, information entropy is used to quantify the logical jumps in the text. First, the foreign language writing assessment content text is windowed, and each window is defined to contain a number of sentences or paragraphs. Then, the information entropy of each window is calculated based on the sentence structure, logical relationship and semantic information in the window. The calculation formula of information entropy is as follows: ,in, Representing terms The probability of occurrence in a paragraph, is the entropy value of the paragraph, The number of words is represented. By analyzing the distribution of each unit in the window, the information entropy of the window is calculated. If the information entropy is high, it means that there is a large information fluctuation in the sentence or paragraph in the window, which is caused by logical jumps. Then, the information entropy of each window is used as the basis for quantifying the interference gradient. By analyzing the changes in each paragraph, the information entropy sequence of the entire text is calculated, and the information entropy sequence interference gradient data is generated. This data contains the entropy value of each paragraph, the corresponding jump type, and the entropy change between paragraphs. This data can be used to quantify the degree of logical jump interference in the text, providing a basis for subsequent native language system interference identification.

[0028] Step S24: performing native language system interference recognition on the foreign language writing test content text according to the information entropy sequence interference gradient data and the content paragraph logical jump data to obtain content paragraph native language system interference recognition data.

[0029] In an embodiment of the present invention, native language system interference recognition is performed based on information entropy sequence interference gradient data and content paragraph logic jump data. The goal of native language system interference recognition is to detect whether there is logical incoherence or improper language structure caused by native language interference in foreign language writing. First, by analyzing the content paragraph logic jump data, combined with the information entropy sequence interference gradient data, it is judged whether the grammatical structure and logical jump in the paragraph meet the standard writing specifications of the target foreign language. By comparing the grammatical and syntactic differences between the native language system and the foreign language system, the language structure abnormality caused by native language interference is identified. For example, the native language tends to use simple sentences in sentence structure, while the target foreign language tends to use complex sentences. In the recognition process, a rule-based interference model is used to detect whether there are features of native language influence in foreign language writing. The model determines which paragraphs or sentences are interfered by the native language by checking the sentence structure, grammatical markers and conjunctions in the text, combined with the fluctuation of information entropy value. Finally, the identified interference information is recorded as content paragraph native language system interference recognition data, including information such as interference type, interference location and interference degree. These data will provide detailed feedback for subsequent foreign language teaching effectiveness evaluation and help identify language structure problems in students' writing.

[0030] Step S23 includes the following steps: Step S231: Count the frequency of logical jump connectives in the foreign language writing assessment content text based on the logical jump data of the content paragraphs to obtain the frequency data of logical jump connectives; Step S232: Decomposing the semantic span word vector of the connective word through the frequency data of the logical jump connective word to obtain the semantic span word vector of the connective word; Step S233: performing a grammatical fuzziness regression analysis of the previous and next paragraphs on the foreign language writing assessment content text according to the semantic span word vector of the connective and the frequency data of the logical jump connective to obtain the grammatical fuzziness regression data of the previous and next paragraphs; Step S234: performing a simulation analysis of the disordered state of the dimensionality constraint of the information entropy sequence based on the semantic span word vector of the connective and the fuzzy regression data of the preceding and following paragraphs, and obtaining the disordered state of the dimensionality constraint of the information entropy sequence; Step S235: quantizing the information entropy sequence interference gradient of the information entropy sequence dimension constraint disorder state to obtain the information entropy sequence interference gradient data.

[0031] In an embodiment of the present invention, in step S231, the frequency statistics of logical jump connectives are performed to analyze the use of connectives in the foreign language writing assessment content text, especially those connectives that cause logical jumps between paragraphs. First, the foreign language writing assessment content text is subjected to lexical and syntactic analysis through syntactic analysis and natural language processing tools (such as SpaCy or NLTK) to extract all connectives. Connectives include but are not limited to "however", "therefore", "because", etc. These words usually appear between paragraphs to indicate the transition of logical relations. Then, by writing a statistical algorithm, the frequency of each connective in the text is calculated and its position is recorded. For each paragraph, the total number of occurrences of the same connective and its relative position in the text are counted to obtain the frequency data of logical jump connectives. These data not only record the frequency of occurrence of each connective, but also include their position in the text, which helps analyze the possibility and intensity of logical jumps in the text. In step S232, the semantic span word vector decomposition of the connective is performed based on the frequency data of the logical jump connective obtained in step S231. First, a word vectorization model (such as Word2Vec or GloVe) is used to learn the semantic representation of each connective. The word vectorization model maps each connective into a multidimensional vector, and each dimension of the vector represents the semantic feature of the word. By analyzing the context of each connective in the text, a word vector is generated, and a vector representation is assigned to each connective. These word vectors can capture the semantic span of the connective, that is, the semantic extension of the connective in a sentence or paragraph. Then, based on the frequency data, the connective is decomposed to generate the semantic span of the word vector. This process requires the word vector to be reduced in dimension by methods such as principal component analysis (PCA) to obtain a more concise and expressive semantic span word vector. Through this process, the semantic influence range of the connective in the text can be quantified, providing key data support for subsequent regression analysis. According to the connective semantic span word vector and the frequency data of the logical jump connective obtained in step S232, a regression analysis of the grammatical ambiguity of the previous and next paragraphs is performed. Grammatical ambiguity refers to the situation in which the text is difficult to understand due to unclear grammar or loose structure. In this step, first, perform grammatical analysis on each paragraph in the text, and use tools based on dependency syntactic analysis (such as SpaCy or Stanford Parser) to identify the syntactic structure of the paragraph. Then, based on the grammatical structure of the previous and next paragraphs, combined with the semantic span and frequency data of the logical jump conjunctions, a regression analysis model (such as linear regression or polynomial regression) is applied to calculate the grammatical ambiguity between paragraphs. Specifically, for each pair of adjacent paragraphs, evaluate whether the logical connection between them is smooth and whether there are abrupt grammatical transitions or structural differences.According to the results of regression analysis, the grammatical fuzzy regression data of the previous and next paragraphs are obtained. The data contains the grammatical fuzziness value between each pair of paragraphs, which indicates the grammatical cohesion between the two paragraphs and the clarity of the grammatical structure. Based on the semantic span word vector of the connective and the grammatical fuzzy regression data of the previous and next paragraphs, the information entropy sequence dimension constraint disorder state simulation analysis is performed. Information entropy is a standard for measuring the disorder and uncertainty of the amount of information in the text. In this step, first, the information entropy sequence of the text is constructed by combining the semantic span word vector obtained in the previous step. This sequence represents the semantic uncertainty within the paragraph by calculating the information entropy value of each paragraph. By defining a sliding window, the grammatical fuzziness and information entropy of the paragraph are calculated in each window, and its dimension is further analyzed. The dimensional constraint indicates the fluctuation range of information entropy in time or space. Using this constraint, the disorder state in the text is simulated, that is, whether there is a sudden logical jump or grammatical inconsistency between paragraphs. To this end, the information entropy formula is used:. ,in, Representing terms The probability of occurrence in a paragraph, is the entropy value of the paragraph, Indicates the number of words and describes the semantic uncertainty within the paragraph. By calculating the entropy value sequence of the entire text, it is identified which parts of the text show a strong disorder state, indicating that there are large logical jumps or unclear grammar between paragraphs. Finally, through model simulation, the dimension-constrained disorder state of the information entropy sequence is obtained, and the entropy value and degree of disorder of each paragraph are recorded to provide data support for subsequent steps. The information entropy sequence interference gradient is quantified for the dimension-constrained disorder state of the information entropy sequence to obtain the information entropy sequence interference gradient data. The purpose of this step is to measure the degree of interference of logical jumps and grammatical ambiguity in the text on the text structure by quantifying the changes in the information entropy sequence. First, using the disorder state obtained in the previous stage, the interference gradient of each paragraph is calculated by analyzing the fluctuations of the information entropy sequence. The interference gradient can be calculated using the following formula: ,in, It's time point The interference gradient at and They represent the information entropy values ​​of the previous and next moments respectively. By calculating the change in information entropy at each moment, the interference gradient between each paragraph is obtained, which indicates the degree of influence caused by logical jumps or grammatical ambiguity between paragraphs. Finally, all interference gradient data are integrated into information entropy sequence interference gradient data, and provide a quantitative basis for subsequent native language system interference identification and foreign language teaching effect evaluation. These interference gradient data can reveal which parts of the writing are subject to strong logical or grammatical interference.

[0032] Step S234 includes the following steps: Perform polysemous clustering analysis on the semantic span word vectors of connectives to obtain polysemous clustering data of connectives; Perform grammatical boundary fuzzy trend analysis on the grammatical fuzzy regression data of the previous and next paragraphs to obtain the grammatical boundary fuzzy trend data of the previous and next paragraphs; Based on the polysemous clustering data of connectives and the fuzzy trend data of the grammatical boundaries of the previous and next paragraphs, the context change boundary fitting is performed to obtain the context change boundary fitting data of the previous and next paragraphs; Based on the Markov chain, the context boundary fitting state is simulated for the context change boundary fitting data before and after the paragraph to obtain the context boundary state fitting data; According to the context boundary state fitting data, the dimension constraint disorder state of the information entropy sequence is simulated and analyzed to obtain the dimension constraint disorder state of the information entropy sequence.

[0033] In an embodiment of the present invention, in step S234, a polysemous clustering analysis of the semantic span word vector of the connective word is performed, provided that the semantic span word vector of the connective word has been obtained through step S232. First, the K-means clustering algorithm is used to perform clustering analysis on the semantic span word vector of the connective word. Specifically, the similarity between the word vectors is measured using the Euclidean distance, and the connective words are classified according to their polysemousness in the context. Through multiple iterations, the K-means algorithm divides the connective words into multiple categories, each of which represents a different semantic representation of the connective word. Then, a label is assigned to each category as "connective polysemous clustering data". For example, if the semantics of the word "because" in different paragraphs are causal relationships, explanations, etc., then the word will be divided into multiple categories in the polysemous clustering. The clustering result generates a polysemous category data for each connective word, which can be used for subsequent context change analysis. Next, the grammatical boundary fuzzy trend analysis is performed on the grammatical fuzzy regression data of the previous and next paragraphs. This analysis aims to reveal whether the grammatical cohesion between paragraphs in the text is gradually blurred. First, based on the grammatical fuzzy regression data of the previous and next paragraphs, the time series analysis method is used to model the changing trend of grammatical fuzziness. A weighted sliding window is used to calculate the fuzziness change of each paragraph, with special attention to the boundary change between paragraphs. By calculating the gradient of the paragraph grammatical fuzziness, the changing trend of the grammatical fuzzy boundary between paragraphs is obtained, namely the "fuzzy trend data of the grammatical boundary between the previous and next paragraphs". This data shows how the connection between paragraphs gradually becomes blurred or clear as the text progresses. Based on the above-mentioned polysemous clustering data of connectives and the fuzzy trend data of the grammatical boundary of the previous and next paragraphs, a context change boundary fitting analysis is performed. This analysis identifies potential context changes in the text by fitting the changes in grammatical fuzziness between paragraphs. First, using the least squares curve fitting technique, the polysemous clustering data of connectives are combined with the fuzzy trend data of the grammatical boundary of paragraphs to construct a model of the context change boundary. By tuning the parameters in the model, the best match between the change of paragraph grammatical fuzziness and the semantic span of connectives is found, thereby obtaining the fitted boundary data of the context change before and after the paragraph. The goal of the fitting process is to minimize the fitting error to obtain the most accurate context change boundary model. This data provides effective boundary data for subsequent context boundary state simulation. Based on the context boundary fitting data, the Markov chain is used to simulate the state fitting of the context change boundary before and after the paragraph. First, the state space of the Markov chain is defined as the possible states of context change between text paragraphs. Each state represents a specific context change boundary. The transfer matrix of the Markov chain is used to represent the probability relationship of context change between paragraphs. By modeling the probability of context change between the previous and next paragraphs, it is possible to predict the impact of context change of a paragraph in the text on the subsequent paragraphs.Using the known context change boundary fitting data, the state transition algorithm of the Markov chain is used for simulation calculation to obtain the context boundary state fitting data in the text. This step further improves the prediction accuracy of context change by simulating the state transition process of context change. Finally, based on the obtained context boundary state fitting data, the information entropy sequence dimension constraint disorder state simulation analysis is performed. This analysis aims to reveal the degree of disorder of the text structure by quantifying the information entropy changes in the text. First, by calculating the context boundary changes between paragraphs, the information entropy formula is used to measure the information uncertainty in the text. The calculation formula of information entropy is as follows:. ,in, Representing terms The probability of occurrence in a paragraph, is the entropy value of the paragraph, The number of words is expressed by combining information entropy with context boundary state fitting data, and using multiple regression analysis or information gain algorithm to further constrain the dimension of entropy value. This step will calculate a disorder value for each paragraph or context state, and finally obtain disorder state data constrained by the dimension of information entropy sequence. This data can reveal the grammatical ambiguity and information entropy fluctuations caused by factors such as context changes and paragraph logic jumps in the text, providing an important basis for the subsequent evaluation of foreign language teaching and teaching effects.

[0034] Step S24 includes the following steps: Step S241: Acquire native language architecture data of different students; Step S242: performing native language migration feature analysis on the foreign language writing assessment content text according to the content paragraph logic jump data and the native language system structure data to obtain the assessment content native language migration feature data; Step S243: Identify syntactic structure logic deviation based on the native language transfer feature data of the test content and the content paragraph logic jump data to obtain native language syntactic structure logic deviation data; Step S244: performing semantic deviation recognition on the native language transfer feature data of the assessment content to obtain native language semantic feature deviation data; Step S245: mapping the foreign language writing assessment content text to a native language dominant syntactic framework according to the native language syntactic structure logic deviation data and the native language ideographic feature deviation data to obtain native language dominant syntactic framework mapping data; Step S246: Perform native language system interference identification based on the information entropy sequence interference gradient data and the native language dominant syntactic framework mapping data to obtain content paragraph native language system interference identification data.

[0035] In the embodiment of the present invention, it is first necessary to obtain the native language system structure data of different students. This data is usually obtained by collecting the language structure information used by students in their native language, which includes but is not limited to vocabulary, syntactic structure, word order, grammatical rules, etc. Through large-scale linguistic data sets, such as the United Nations Language Dataset or various known native language annotated corpora, the core syntactic structure features of each student's native language are extracted. In specific implementation, the native language text provided by each student is syntactically analyzed using a dependency syntactic analysis algorithm. The analysis tool can use, for example, Stanford Parser, which is used to parse the hierarchical structure of sentence components, extract various syntactic components (such as subject, predicate, object, modifying components, etc.), and record the position and frequency of each component in the sentence. By extracting and encoding these syntactic elements, the structural data of each student's native language system, such as different sentence patterns, common grammatical rules, etc., are obtained. This data provides a basis for subsequent native language migration feature analysis. In step S242, the native language migration feature analysis of the foreign language writing assessment content is performed based on the content paragraph logic jump data and the native language system structure data. First, the paragraph logic jump data obtained above is combined with the mother tongue system data of each student to analyze whether the grammatical structure in the foreign language writing is affected by the mother tongue structure. In the specific implementation, a pattern matching algorithm is used to identify the structural similarity between the grammatical structure in the foreign language writing text and the mother tongue system. For example, when the subject-verb-object structure of the foreign language sentence is similar to the common structure of the mother tongue sentence, it can be considered that this is one of the characteristics of mother tongue migration. By comparing the difference between the sentence structure in each student's foreign language writing text and the common pattern of its mother tongue system, the potential influence of the mother tongue on foreign language writing is identified, and "mother tongue migration feature data" is generated. These data identify which grammatical structures are affected by the mother tongue, further revealing the mother tongue migration phenomenon in the student's writing. Step S243 involves the identification of syntactic structure logic deviation of the foreign language writing assessment content based on the mother tongue migration feature data and paragraph logic jump data. In this step, it is first necessary to analyze the grammatical deviation in the foreign language writing, that is, the syntactic structure errors or logical inconsistencies that occur in the student's foreign language writing. By comparing the native language migration feature data with the syntactic structure in foreign language writing, syntactic analysis tools (such as NLTK, spaCy, etc.) can be used to perform structured analysis on the sentences in the writing and detect the syntactic deviations therein. For the identification of deviations, the matching and comparison method of the grammatical tree is adopted. For example, by calculating the similarity of the syntactic tree, it is judged whether there is a syntactic structure that does not conform to the foreign language rules. In addition, by introducing a logical deviation analysis method (such as logistic regression or decision tree), the logical jump deviation between sentences is identified according to the logical relationship between paragraphs. These steps ultimately result in "native language syntactic structure logical deviation data", which indicates the syntactic logic inconsistency caused by the influence of the native language in foreign language writing. In step S244, semantic deviation identification is performed.The goal of semantic deviation identification is to analyze the expression errors caused by mother tongue transfer in foreign language writing, especially the meaning communication deviation that occurs when students use their mother tongue thinking framework to construct foreign language sentences. To this end, semantic analysis technology (such as a word vector model based on Word2Vec or GloVe) is used to analyze foreign language writing at the semantic level. By comparing the semantic associations between words in the student writing text and their mother tongue, the parts with miscommunication of meaning are identified. For example, when the semantic span of some words in the mother tongue and the foreign language is very different, semantic deviation is caused. The degree of semantic deviation of each word is calculated using an algorithm based on semantic similarity (such as cosine similarity), and the degree of semantic deviation of each paragraph is quantified by regression analysis. These calculations result in "mother tongue semantic feature deviation data", which reflects the meaning communication problems caused by the influence of the mother tongue in foreign language writing. In step S245, the mother tongue dominant syntactic framework mapping is performed in combination with the mother tongue syntactic structure logic deviation data and the mother tongue semantic feature deviation data. This step creates a mapping model by comprehensively analyzing the syntactic structure and semantic deviation in the foreign language text, and matches the syntactic structure in the foreign language text with the syntactic framework of the native language. The specific method adopted is to fit the relationship between the native syntactic framework and the syntactic structure in the foreign language writing based on a mapping algorithm (such as linear regression, polynomial regression, etc.). Through this step, it is possible to identify which parts of the foreign language text are completely consistent with the syntactic structure of the native language and which parts deviate from the logic of the native language. Finally, the "native language dominant syntactic framework mapping data" is obtained to show how the foreign language sentences in the students' writing are affected by the native language structure, and to represent the adaptation of the foreign language sentences to the native language syntactic structure in a graphical or numerical way. In step S246, native language system interference identification is performed according to the information entropy sequence interference gradient data and the native language dominant syntactic framework mapping data. This step combines the information entropy sequence interference gradient data with the native language dominant syntactic framework mapping data to further analyze how the native language system affects the overall structure of foreign language writing. First, by using information entropy to quantify the degree of interference in the text, the interference gradient data obtained previously was used to conduct interference analysis on foreign language writing, revealing the interference areas of the native language in foreign language writing. Then, combined with the native language dominant syntactic framework mapping data, clustering analysis methods (such as K-means clustering) were used to cluster foreign language writing texts to identify the sentence areas most significantly affected by the native language. These analyses ultimately generated "content paragraph native language system interference identification data" to identify potential error areas in foreign language writing caused by native language interference and provide a basis for subsequent teaching evaluation.

[0036] Step S3 includes the following steps: Step S31: Obtain the underlying learning architecture of the AI ​​tool; Step S32: normalizing the information entropy sequence interference gradient data to obtain information entropy sequence interference gradient normalized data; Step S33: Perform AI architecture disorder degree recognition learning reinforcement on the underlying learning architecture of the AI ​​tool according to the information entropy sequence interference gradient normalization data to obtain an AI disorder degree recognition learning reinforcement architecture; Step S34: Based on the content paragraph native language system interference recognition data, the AI ​​disorder degree recognition learning reinforcement architecture is optimized for adaptive language system anomaly recognition to obtain the AI ​​language system anomaly recognition architecture.

[0037] As an example of the present invention, refer to Figure 3 As shown, in this example, step S3 includes: Step S31: Obtain the underlying learning architecture of the AI ​​tool; In the embodiment of the present invention, it is first necessary to obtain the underlying learning architecture of the AI ​​tool. The underlying learning architecture of the AI ​​tool is usually composed of a neural network, a deep learning framework, an optimization algorithm, etc. In the specific implementation, the underlying learning architecture uses a deep neural network (DNN) or a convolutional neural network (CNN) as the basic structure to process the input data and perform learning and prediction. The architecture includes multiple neural network layers, each layer contains a different number of neurons, and performs nonlinear transformations through activation functions (such as ReLU, Sigmoid). Depending on the task, the learning architecture also includes a recurrent neural network (RNN) for processing time series data, or a long short-term memory network (LSTM) for processing long-term dependencies. The design of the underlying learning architecture needs to be optimized according to the specific task, and usually the hyperparameters of the training model, such as the learning rate, batch size, optimizer type (such as Adam or SGD), etc., are selected to ensure that the model can be effectively trained on a given data set.

[0038] Step S32: normalizing the information entropy sequence interference gradient data to obtain information entropy sequence interference gradient normalized data; In an embodiment of the present invention, the information entropy sequence interference gradient data is normalized to obtain the information entropy sequence interference gradient normalized data. The information entropy sequence interference gradient data represents the uncertainty and interference degree in the sequence, and the purpose of the normalization process is to scale the data to a standard range to eliminate the influence of different data ranges on subsequent analysis. Specifically, a standardization method (such as Z-score normalization or Min-Max normalization) is used to process the data.

[0039] Step S33: Perform AI architecture disorder degree recognition learning reinforcement on the underlying learning architecture of the AI ​​tool according to the information entropy sequence interference gradient normalization data to obtain an AI disorder degree recognition learning reinforcement architecture; In an embodiment of the present invention, the underlying learning architecture of the AI ​​tool is reinforced for the disorder degree recognition learning of the AI ​​architecture according to the information entropy sequence interference gradient normalized data. The core goal of this step is to use the normalized interference gradient data to enhance the disorder degree recognition ability of the AI ​​architecture. Specifically, it is first necessary to adjust the parameters (such as the weight matrix) in the AI ​​model so that it can better recognize the disordered patterns in the data. The network is trained by using the backpropagation algorithm (Backpropagation) to adjust the weights in the network to minimize the loss function of disorder degree recognition. The loss function can be defined based on information entropy or other uncertainty measures, and the weights of the network are adjusted by the backpropagation optimization algorithm (such as the gradient descent method) until the network can accurately recognize the disordered patterns in the data, thereby obtaining the enhanced AI architecture. In this process, the AI ​​model not only trains the prediction target, but also strengthens the recognition ability of disordered patterns so that it can maintain efficient learning ability in the face of complex input data.

[0040] Step S34: Based on the content paragraph native language system interference recognition data, the AI ​​disorder degree recognition learning reinforcement architecture is optimized for adaptive language system anomaly recognition to obtain the AI ​​language system anomaly recognition architecture.

[0041] In an embodiment of the present invention, the AI ​​disorder degree recognition learning reinforcement architecture is optimized for adaptive language system anomaly recognition based on the native language system interference recognition data of the content paragraph. The goal of this step is to further optimize the architecture of the AI ​​model through the native language interference recognition data, so that it can more effectively identify and handle anomalies in different language systems. First, the native language system interference recognition data is used as additional input information to optimize the AI ​​model through an adaptive learning algorithm (such as adaptive gradient descent). In this process, the native language interference data is regarded as a kind of noise or interference, and the AI ​​architecture needs to learn how to extract effective information from these interferences. The specific operation method is to dynamically adjust the parameters in the learning process by adjusting the learning rate and the optimization algorithm, so that the AI ​​architecture can better adapt to the abnormal patterns brought by different language systems. The loss function used by the optimization algorithm is in the form of: ,in, is the original loss function, which is the loss calculated by the AI ​​model based on the input data, usually related to the difference between the predicted results and the true label (for example, cross entropy loss or mean squared error), is the regularization coefficient, which is used to balance the impact of the original loss function and the native language interference term. represents the sum of the native language interference recognition errors, where: is the true value of the native language interference data, data points of native language interference markers (which can be actual interference patterns, indicating the influence of the native language on foreign language writing), For AI models The predicted value of the data point is the native language interference predicted by the model. is the number of data points, that is, how many native language interference data are used to train the model. Finally, based on this optimization process, the resulting "AI language system anomaly recognition architecture" can more accurately identify abnormal language patterns caused by native language interference, thereby effectively optimizing the recognition ability of AI tools when processing foreign language writing tasks.

[0042] Step S33 includes the following steps: Step S331: performing topological information entropy discretization processing on the information entropy sequence interference gradient normalized data to obtain an information entropy topological discrete feature space; Step S332: performing feature orthogonal mapping and information entropy reconstruction processing on the information entropy topological discrete feature space to obtain a feature mapping information entropy tensor; Step S333: performing interference gradient calibration processing according to the feature mapping information entropy tensor to obtain interference gradient calibration data; Step S334: Based on the interference gradient calibration data, the underlying learning architecture of the AI ​​tool is subjected to AI architecture disorder degree recognition learning reinforcement to obtain an AI disorder degree recognition learning reinforcement architecture.

[0043] In an embodiment of the present invention, the input information entropy sequence interference gradient normalized data is first subjected to topological information entropy discretization processing. Specifically, the information entropy sequence interference gradient normalized data is represented as an ordered data set, which reflects the gradient change of the data and its distribution state under the information entropy metric. The topological information entropy discretization process discretizes the topological structure of these data and divides them into a finite number of discrete intervals in the continuous space. This process uses the Voronoi diagram for discretization, divides the data space into multiple non-overlapping regions, and represents the characteristics of the information entropy in each region. Through this discretization process, the originally continuous interference gradient data can be mapped to a discrete feature space, which can better express the topological relationship implicit in the data. The discretized data is the information entropy topological discrete feature space, and the coordinates of each point correspond to the characteristic value in a specific region, which is convenient for the subsequent processing and analysis of the information entropy feature. Using the feature orthogonal mapping method, the information entropy topological discrete feature space obtained in step S331 is mapped. First, the principal component analysis (PCA) algorithm is used to perform feature orthogonal mapping to orthogonalize the relevant features in the feature space to ensure that the features are independent of each other and avoid multicollinearity problems. In this process, the eigenvalues ​​and eigenvectors are calculated by constructing the covariance matrix to obtain the mapped eigenvector space. Next, information entropy reconstruction is performed, that is, the information entropy data is reconstructed through methods such as Fourier transform or wavelet transform to retain its original spectral characteristics. This process converts the information entropy features in the feature space into a high-dimensional tensor structure, so that each feature dimension can express more information and is more suitable for complex multidimensional data modeling. The feature mapping information entropy tensor finally obtained not only retains the global characteristics of information entropy, but also introduces the orthogonal relationship between each feature, providing a more accurate representation for subsequent data calibration and learning. Interference gradient calibration processing is performed based on the feature mapping information entropy tensor. The specific operation is to first calculate the gradient of the mapped tensor by the gradient descent method to detect the influence of interference factors in the current feature space on model training. This process uses the back propagation algorithm to calculate the error in the model and perform gradient updates. At this time, the output of the model will be dynamically adjusted according to the information entropy characteristics in each dimension to gradually reduce the influence of unnecessary interference. By gradually reducing the interference, the interference gradient is finally calibrated. During the calibration process, the Adam optimization algorithm is used to automatically adjust the learning rate to speed up the convergence speed and ensure that the interference gradient is effectively corrected. The calibrated data is the interference gradient calibration data. The core of this step is to ensure that the model responds more accurately in each feature dimension through an adaptive optimization algorithm, reduce the impact of interference signals, and enable the model to better learn the target task. Based on the interference gradient calibration data obtained in step S333, the underlying learning architecture of the AI ​​tool is reinforced. Specifically, the reinforcement learning method is used to gradually improve the learning process of the AI ​​model through environmental interaction.At this point, the AI ​​model learns how to maintain high learning efficiency and accuracy in the face of interfering data by identifying the degree of disorder in the training data (i.e., inconsistencies and abnormal patterns in the data) and adjusting its learning strategy. In this process, the Q-learning algorithm is used to identify the degree of disorder, and the model continuously optimizes its parameters by evaluating the reward value of each operation. Through reinforcement learning, the AI ​​architecture is gradually able to adapt to different degrees of disordered data and can adjust itself in new data environments to achieve stronger learning adaptability. The resulting AI disorder degree recognition learning reinforcement architecture can better cope with various data disturbances and improve the stability and accuracy of the model in complex environments.

[0044] Step S34 includes the following steps: Step S341: performing native language part-of-speech interference anomaly logic analysis on the native language system interference recognition data of the content paragraph to obtain native language part-of-speech interference anomaly logic data; Step S342: Summarizing the part-of-speech logic metaphor deviation based on the native language part-of-speech interference abnormal logic data to obtain the part-of-speech logic metaphor deviation data; Step S343: performing concept cross-domain metaphor anomaly analysis on the part-of-speech logic metaphor deviation data to obtain concept cross-domain metaphor anomaly data; Step S344: Based on the part-of-speech logic metaphor deviation data and the concept cross-domain metaphor anomaly data, the AI ​​disorder degree recognition learning reinforcement architecture is adaptively optimized for language system anomaly recognition to obtain the AI ​​language system anomaly recognition architecture.

[0045] In an embodiment of the present invention, the mother tongue system interference identification data of the content paragraph is processed, and the mother tongue part-of-speech interference abnormal logic analysis is performed to identify the irregular or inconsistent use of parts of speech in foreign language writing due to mother tongue interference. Specifically, the mother tongue system interference data includes the part of foreign language writing that is affected by the mother tongue, especially at the level of vocabulary and grammatical structure. Through syntactic analysis and dependency syntactic analysis, combined with linguistic part-of-speech tagging, these data are logically analyzed. The conditional random field (CRF) algorithm is used to refine the part-of-speech tagging to find out those part-of-speech conversions or incorrect tags that do not conform to the target foreign language rules. Then, combined with the language model, the abnormal logical patterns are identified by calculating the probability of each part-of-speech conversion. These abnormal patterns are usually manifested as the misuse of parts of speech under the influence of the mother tongue in certain contexts. Through this analysis, the mother tongue part-of-speech interference abnormal logic data is obtained, which represents the regularity and abnormal performance of part-of-speech interference in foreign language writing affected by the mother tongue, and provides key data for subsequent deviation induction and abnormal optimization. According to the abnormal logic data of native language part-of-speech interference obtained in step S341, the part-of-speech logic metaphor deviation is summarized. The purpose of summarizing the part-of-speech logic metaphor deviation is to identify the implicit deviation of the use of part-of-speech in the target foreign language due to the transformation of the native language structure or expression. Through the metaphor analysis method, the metaphorical expressions in the text are first identified, especially those metaphorical errors caused by the change of part-of-speech leading to the conversion of meaning. For example, some words have a certain grammatical function in the native language, but do not have the same function in the foreign language, or their part-of-speech conversion leads to a significant deviation in the meaning of the sentence. Then, the reasoning algorithm is used to summarize and analyze these metaphorical deviations, classify them into specific types of logical errors, and further refine the analysis according to the type of part-of-speech conversion. The summarized metaphorical deviations are classified and processed using the decision tree algorithm to obtain specific part-of-speech logic metaphor deviation data, which provides quantitative analysis for metaphor conversion errors of different parts of speech, and can accurately show the potential error types and conditions under the influence of the native language. Based on the part-of-speech logic metaphor deviation data obtained in step S342, the concept cross-domain metaphor anomaly analysis is further performed. Conceptual cross-domain metaphor refers to the situation in which, in foreign language writing, due to the influence of the native language system, the use of parts of speech or concepts exceeds their original context or domain, resulting in abnormal or illogical language expression. For example, when translating, some concepts in the native language are directly applied to the foreign language, and this cross-domain metaphor will lead to semantic inconsistency. By expanding metaphor identification and combining semantic network analysis methods, based on dictionary resources such as WordNet, the reasonable meaning of words in different contexts is determined, and those cross-domain conceptual metaphors that do not conform to foreign language usage habits are identified. Hierarchical clustering analysis is used to classify metaphorical anomalies according to their cross-domain degree, with special attention paid to those parts of speech usage problems caused by concept transfer.The final concept cross-domain metaphor anomaly data can accurately capture the cross-domain metaphor deviations and abnormal patterns that appear in foreign languages ​​due to native language interference, and demonstrate the cross-domain metaphor deviations and abnormal patterns that appear in different domains at the language level and the implicit impact they produce. Based on the part-of-speech logic metaphor deviation data and concept cross-domain metaphor abnormality data obtained in steps S342 and S343, the AI ​​disorder degree recognition learning reinforcement architecture is adaptively optimized for language system abnormality recognition. At this time, the AI ​​architecture needs to be adaptively optimized for various abnormal patterns that appear in the foreign language learning process. First, these two types of data are used as training inputs, and data fusion processing is performed through the convolutional neural network (CNN) in deep learning. Combined with the long short-term memory network (LSTM) model, through time series modeling, the dynamic changes of metaphor deviations and cross-domain metaphor anomalies are captured, and the weight distribution in the learning process is optimized. Then, through the Q-learning method in reinforcement learning, the AI ​​architecture gradually improves the ability to recognize language system anomalies based on the reward and punishment mechanism for misidentification. In this way, the AI ​​architecture can automatically adjust its parameters during the training process and enhance the ability to recognize implicit anomalies in the language system. Ultimately, the resulting AI language system anomaly recognition architecture can adjust the model’s learning path in real time, more efficiently identify abnormal patterns in the language, and improve the accuracy of foreign language writing assessment.

[0046] The present invention also provides an AI-based foreign language teaching effect evaluation system, which is used to execute the AI-based foreign language teaching effect evaluation method as described above. The AI-based foreign language teaching effect evaluation system includes: The paragraph logic segmentation module is used to obtain the foreign language writing assessment content text submitted by the students; perform paragraph logic segmentation on the foreign language writing assessment content text to obtain paragraph logic segmentation data of the foreign language writing assessment content; The interference recognition module is used to quantify the information entropy sequence interference gradient of the foreign language writing assessment content text based on the paragraph logic segmentation data to obtain the information entropy sequence interference gradient data; perform native language system interference recognition on the foreign language writing assessment content text based on the information entropy sequence interference gradient data to obtain the content paragraph native language system interference recognition data; The anomaly recognition optimization module is used to obtain the underlying learning architecture of the AI ​​tool; the underlying learning architecture of the AI ​​tool is optimized for adaptive language system anomaly recognition based on the native language system interference recognition data of the content paragraphs to obtain the AI ​​language system anomaly recognition architecture; The effect evaluation module is used to evaluate the foreign language teaching effect on the foreign language writing assessment content text according to the AI ​​language system anomaly recognition architecture, obtain the foreign language teaching effect evaluation data, and send the foreign language teaching effect evaluation data to the terminal.

[0047] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A foreign language teaching effect evaluation method based on AI, characterized in that: The following steps are involved: Step S1: obtaining the foreign language writing assessment content text submitted by the student; performing paragraph logical segmentation on the foreign language writing assessment content text to obtain paragraph logical segmentation data of the foreign language writing assessment content; Step S2: quantifying the information entropy sequence interference gradient of the foreign language writing test content text based on the paragraph logic segmentation data to obtain the information entropy sequence interference gradient data; According to the information entropy sequence interference gradient data, native language system interference recognition is performed on the foreign language writing test content text to obtain the content paragraph native language system interference recognition data; Step S3: Obtain the underlying learning architecture of the AI ​​tool; perform adaptive language system anomaly recognition optimization on the underlying learning architecture of the AI ​​tool according to the native language system interference recognition data of the content paragraphs, and obtain the AI ​​language system anomaly recognition architecture; Step S4: Perform foreign language teaching effectiveness evaluation on the foreign language writing assessment content text according to the AI ​​language system anomaly recognition architecture, obtain foreign language teaching effectiveness evaluation data, and send the foreign language teaching effectiveness evaluation data to the terminal.

2. The AI-based foreign language teaching effect evaluation method according to claim 1 is characterized in that: Step S1 includes the following steps: Step S11: Obtaining the foreign language writing test content text submitted by the student; Step S12: cleaning the foreign language writing assessment content text to obtain a cleaned foreign language writing assessment content text; Step S13: Performing a chapter structure analysis on the cleaned text of the foreign language writing assessment content to obtain the chapter structure data of the writing content; Step S14: logically segment the cleaned text of the foreign language writing assessment content according to the chapter structure narrative data of the writing content to obtain logical segmentation data of the foreign language writing assessment content.

3. The AI-based foreign language teaching effect evaluation method according to claim 1 is characterized in that: Step S2 includes the following steps: Step S21: performing sentence parallel / subordinate connection analysis on the foreign language writing assessment content text according to the paragraph logic segmentation data to obtain sentence parallel / subordinate connection data; Step S22: Based on the sentence parallel / subordinate connection data and the paragraph logic segmentation data, the foreign language writing test content text is identified with logic jump links to obtain content paragraph logic jump data; Step S23: quantifying the information entropy sequence interference gradient of the foreign language writing test content text based on the content paragraph logic jump data to obtain information entropy sequence interference gradient data; Step S24: performing native language system interference recognition on the foreign language writing test content text according to the information entropy sequence interference gradient data and the content paragraph logical jump data to obtain content paragraph native language system interference recognition data.

4. The AI-based foreign language teaching effect evaluation method according to claim 3 is characterized in that: Step S23 includes the following steps: Step S231: Count the frequency of logical jump connectives in the foreign language writing assessment content text based on the logical jump data of the content paragraphs to obtain the frequency data of logical jump connectives; Step S232: Decomposing the semantic span word vector of the connective word through the frequency data of the logical jump connective word to obtain the semantic span word vector of the connective word; Step S233: performing a grammatical fuzziness regression analysis of the previous and next paragraphs on the foreign language writing assessment content text according to the semantic span word vector of the connective and the frequency data of the logical jump connective to obtain the grammatical fuzziness regression data of the previous and next paragraphs; Step S234: performing a simulation analysis of the disordered state of the dimensionality constraint of the information entropy sequence based on the semantic span word vector of the connective and the fuzzy regression data of the preceding and following paragraphs, and obtaining the disordered state of the dimensionality constraint of the information entropy sequence; Step S235: quantizing the information entropy sequence interference gradient of the information entropy sequence dimension constraint disorder state to obtain the information entropy sequence interference gradient data.

5. The AI-based foreign language teaching effect evaluation method according to claim 4 is characterized in that: Step S234 includes the following steps: Perform polysemous clustering analysis on the semantic span word vectors of connectives to obtain polysemous clustering data of connectives; Perform grammatical boundary fuzzy trend analysis on the grammatical fuzzy regression data of the previous and next paragraphs to obtain the grammatical boundary fuzzy trend data of the previous and next paragraphs; Based on the polysemous clustering data of connectives and the fuzzy trend data of the grammatical boundaries of the previous and next paragraphs, the context change boundary fitting is performed to obtain the context change boundary fitting data of the previous and next paragraphs; Based on the Markov chain, the context boundary fitting state is simulated for the context change boundary fitting data before and after the paragraph to obtain the context boundary state fitting data; According to the context boundary state fitting data, the dimension constraint disorder state of the information entropy sequence is simulated and analyzed to obtain the dimension constraint disorder state of the information entropy sequence.

6. The AI-based foreign language teaching effect evaluation method according to claim 3 is characterized in that: Step S24 includes the following steps: Step S241: Acquire native language architecture data of different students; Step S242: performing native language migration feature analysis on the foreign language writing assessment content text according to the content paragraph logic jump data and the native language system structure data to obtain the assessment content native language migration feature data; Step S243: Identify syntactic structure logic deviation based on the native language transfer feature data of the test content and the content paragraph logic jump data to obtain native language syntactic structure logic deviation data; Step S244: performing semantic deviation recognition on the native language transfer feature data of the assessment content to obtain native language semantic feature deviation data; Step S245: mapping the foreign language writing assessment content text to a native language dominant syntactic framework according to the native language syntactic structure logic deviation data and the native language ideographic feature deviation data to obtain native language dominant syntactic framework mapping data; Step S246: Perform native language system interference identification based on the information entropy sequence interference gradient data and the native language dominant syntactic framework mapping data to obtain content paragraph native language system interference identification data.

7. The AI-based foreign language teaching effect evaluation method according to claim 1 is characterized in that: Step S3 includes the following steps: Step S31: Obtain the underlying learning architecture of the AI ​​tool; Step S32: normalizing the information entropy sequence interference gradient data to obtain information entropy sequence interference gradient normalized data; Step S33: Perform AI architecture disorder degree recognition learning reinforcement on the underlying learning architecture of the AI ​​tool according to the information entropy sequence interference gradient normalization data to obtain an AI disorder degree recognition learning reinforcement architecture; Step S34: Based on the content paragraph native language system interference recognition data, the AI ​​disorder degree recognition learning reinforcement architecture is optimized for adaptive language system anomaly recognition to obtain the AI ​​language system anomaly recognition architecture.

8. The AI-based foreign language teaching effect evaluation method according to claim 7 is characterized in that: Step S33 includes the following steps: Step S331: performing topological information entropy discretization processing on the information entropy sequence interference gradient normalized data to obtain an information entropy topological discrete feature space; Step S332: performing feature orthogonal mapping and information entropy reconstruction processing on the information entropy topological discrete feature space to obtain a feature mapping information entropy tensor; Step S333: performing interference gradient calibration processing according to the feature mapping information entropy tensor to obtain interference gradient calibration data; Step S334: Based on the interference gradient calibration data, the underlying learning architecture of the AI ​​tool is subjected to AI architecture disorder degree recognition learning reinforcement to obtain an AI disorder degree recognition learning reinforcement architecture.

9. The AI-based foreign language teaching effect evaluation method according to claim 7 is characterized in that: Step S34 includes the following steps: Step S341: performing native language part-of-speech interference anomaly logic analysis on the native language system interference recognition data of the content paragraph to obtain native language part-of-speech interference anomaly logic data; Step S342: Summarizing the part-of-speech logic metaphor deviation based on the native language part-of-speech interference abnormal logic data to obtain the part-of-speech logic metaphor deviation data; Step S343: performing concept cross-domain metaphor anomaly analysis on the part-of-speech logic metaphor deviation data to obtain concept cross-domain metaphor anomaly data; Step S344: Based on the part-of-speech logic metaphor deviation data and the concept cross-domain metaphor anomaly data, the AI ​​disorder degree recognition learning reinforcement architecture is adaptively optimized for language system anomaly recognition to obtain the AI ​​language system anomaly recognition architecture.

10. An AI-based foreign language teaching effect evaluation system, characterized in that: Used to execute the AI-based foreign language teaching effect evaluation method as claimed in claim 1, the AI-based foreign language teaching effect evaluation system comprises: The paragraph logic segmentation module is used to obtain the foreign language writing assessment content text submitted by the students; perform paragraph logic segmentation on the foreign language writing assessment content text to obtain paragraph logic segmentation data of the foreign language writing assessment content; The interference recognition module is used to quantify the information entropy sequence interference gradient of the foreign language writing assessment content text based on the paragraph logic segmentation data to obtain the information entropy sequence interference gradient data; perform native language system interference recognition on the foreign language writing assessment content text based on the information entropy sequence interference gradient data to obtain the content paragraph native language system interference recognition data; The anomaly recognition optimization module is used to obtain the underlying learning architecture of the AI ​​tool; the underlying learning architecture of the AI ​​tool is optimized for adaptive language system anomaly recognition based on the native language system interference recognition data of the content paragraphs to obtain the AI ​​language system anomaly recognition architecture; The effect evaluation module is used to evaluate the foreign language teaching effect on the foreign language writing assessment content text according to the AI ​​language system anomaly recognition architecture, obtain the foreign language teaching effect evaluation data, and send the foreign language teaching effect evaluation data to the terminal.

Images