A method and system for processing non-standardized interference in a job

Abstract

This invention relates to a method and system for handling non-standardized interference in homework assignments. The method includes: receiving homework data to be processed, the homework data including at least text data and / or image data; processing the homework data using a perceptual model to extract perceptual features of the homework data; processing the homework data using a cognitive model based on the perceptual features of the homework data; fusing perceptual and cognitive features to identify the types of non-standardized interference present in the homework data; and formulating and executing corresponding processing strategies according to the identified types of non-standardized interference to generate an analysis report containing interference correction suggestions or content clarification. By understanding the visual form and semantic content of homework assignments through perceptual and cognitive models, the accuracy of grading is improved when faced with non-standardized interference such as illegible handwriting and disordered formatting. It can automatically process massive amounts of homework, greatly improving grading efficiency and reducing the burden on teachers.

Description

Technical Field

[0001] This invention relates to the fields of educational informatization and artificial intelligence technology, and in particular to a method and system for handling non-standardized interference in homework assignments. Background Technology

[0002] In today's education field, student homework correction is a crucial part of the teaching process. It plays a key role in helping teachers understand students' learning progress, assess teaching effectiveness, and provide targeted guidance. However, current student homework correction faces numerous non-standardized interference issues, which seriously affect the accuracy, efficiency, and overall quality of correction.

[0003] Sloppy handwriting is one of the most common problems. With increasing academic workload and the difficulty in balancing writing speed and quality, many students' homework assignments are illegible. In a survey of primary and secondary school students, more than half of the students' homework contained sloppy handwriting. Some students use excessive cursive writing and overlapping strokes, making the answer area look like a jumbled mess after scanning. Teachers have to grade hundreds or even thousands of papers every day, with only a few dozen seconds to grade each paper on average, leaving them no time to carefully decipher these illegible characters. In electronic grading, this sloppy handwriting becomes even more difficult to read after scanning. According to statistics from a provincial college entrance examination grading team, papers with misjudged handwriting account for 3%-5% of the total number of papers graded each year, seriously affecting students' grades.

[0004] The phenomenon of non-standard answering is also common. Students often fail to follow established answer formats and requirements when solving problems. For example, in math homework, the solution steps are disorganized, and important reasoning processes are missing or skipped; in Chinese composition, paragraph divisions are unclear, and punctuation is used arbitrarily. In a math exam grading session, it was found that about 30% of students had incomplete steps and incoherent logic in their geometry proofs, resulting in significant point deductions. This not only reflects students' insufficient grasp of knowledge but also their lack of good answering habits, making it difficult for teachers to accurately assess students' knowledge level and thought processes during grading.

[0005] Vague language expression also poses a significant challenge to homework grading. Students often fail to articulate their views clearly and accurately when expressing themselves or answering questions, resulting in semantic ambiguity and logical inconsistencies. In grading English writing assignments, frequent grammatical errors, inappropriate vocabulary use, and disorganized sentence structures frequently arise, making the content difficult to understand. This forces teachers to spend considerable time deciphering students' intentions, increasing the difficulty and workload of grading and hindering the ability to provide precise feedback and guidance.

[0006] Traditional methods of grading student assignments primarily rely on manual teacher marking. However, these methods prove inadequate and have significant limitations when faced with non-standardized errors. Manual marking struggles to guarantee accuracy and objectivity, especially with large volumes of assignments. Teachers are prone to fatigue and oversight, leading to misjudgments of illegible handwriting, non-standard answers, and unclear language. Furthermore, manual marking is inefficient, requiring teachers to spend considerable time and energy grading each assignment individually. This results in excessively long feedback times, preventing students from receiving timely guidance and negatively impacting their learning. Additionally, manual marking makes it difficult to comprehensively and deeply analyze non-standardized issues in student assignments, failing to provide robust data support for improving teaching.

[0007] In conclusion, the current problem of non-standardized interference in student homework grading urgently requires effective solutions, as traditional methods can no longer meet the needs of modern education. Therefore, researching a method and system for handling non-standardized interference in student homework based on the fusion of perception and cognitive models is of significant practical importance and urgency. It can significantly improve the accuracy, efficiency, and intelligence of homework grading, providing strong support for education and teaching. Summary of the Invention

[0008] The technical problem to be solved by the present invention is to provide a method and system for processing non-standardized interference in homework based on the fusion of perception and cognitive models. By constructing a deep fusion perception model and cognitive model, the method achieves collaborative processing and deep understanding of homework image and text information, thereby accurately identifying and processing non-standardized interference such as illegible handwriting, format errors, and semantic ambiguity, and improving the accuracy, efficiency and intelligence level of homework correction.

[0009] Firstly, the technical solution adopted by this invention is a method for handling non-standardized operational interference, which includes the following steps: S1. Receive job data to be processed, wherein the job data includes at least text data and / or image data; S2. Process the job data using a perception model to extract the perception features of the job data. Features include text content features and / or layout and graphic features extracted through image recognition; S3. Based on the perceptual features of the homework data, the homework data is processed through a cognitive model to extract the cognitive features of the homework data. The cognitive features include semantic understanding features extracted based on natural language processing (NLP) and / or problem-solving logic features extracted based on logical reasoning. S4. By fusing the perceptual features and the cognitive features, the non-standardized interference types present in the job data are identified. S5. Based on the identified non-standardized interference types, formulate and implement corresponding processing strategies to generate an analysis report that includes interference correction suggestions or content clarification.

[0010] The beneficial effects of this invention are as follows: By employing the aforementioned method for handling non-standardized interference in assignments, and through perceptual and cognitive models to understand the visual form and semantic content of assignments, the accuracy of grading is improved when faced with non-standardized interference such as illegible handwriting and disordered formatting. This method can automatically process massive amounts of assignments, identify various complex interference types, and generate targeted processing suggestions, greatly improving grading efficiency and reducing the burden on teachers. The in-depth analysis report output by this method provides teachers with accurate learning data and students with personalized feedback, achieving the goals of precise teaching and personalized learning.

[0011] Preferably, in step S2, the processing of the job data through the perception model specifically includes: If the job data includes image data, the image data is preprocessed, including grayscale conversion, etc. At least one of binarization, denoising, and tilt correction is performed; based on optical character recognition (OCR) technology, text recognition is performed on the preprocessed image data to extract text content and its... Location information is used as a feature of the text content; based on image recognition technology, the preprocessed image data is analyzed to locate the areas containing the questions and the answers. Multiple structured areas, including the chart area, are extracted, and layout features describing the location, size, and relative relationship of each area are extracted. Non-textual visual elements are identified and extracted from the preprocessed image data to obtain graphic features that describe graphic categories, geometric properties, or chart structures. If the job data contains text data, the text data is encoded and standardized preprocessed, and the preprocessed data is then... The processed text data serves as the text content feature.

[0012] Preferably, in step S3, the processing of the task data using a cognitive model specifically includes: Based on the text content features, a natural language processing (NLP) model is used to perform lexical, syntactic, and deep semantic analysis, extract the semantic representation vectors and key semantic relationships of the text, and generate the semantic understanding features. Based on the text content features, layout features, and graphic features, the task content is transformed into a structured representation. A preset domain knowledge base and reasoning rules are invoked to perform logical verification and problem-solving path analysis, generating the problem-solving logic features that describe the problem-solving approach.

[0013] Preferably, in step S4, the fusion of the perceptual features and the cognitive features specifically includes at least one of the following methods: data layer fusion, feature layer fusion, and decision layer fusion; wherein... Data layer fusion involves associating and aligning the original task image data with the text content features to form multimodal fusion data; feature layer fusion involves fusing at least two of the text content features, layout features, graphic features, semantic understanding features, and problem-solving logic features through a neural network to generate a joint feature representation; decision layer fusion involves obtaining a first judgment about the surface morphology of the task based on the perceptual features, obtaining a second judgment about the quality of the task content based on the cognitive features, and making a comprehensive decision based on the first and second judgments to obtain the identification result of the non-standardized interference type.

[0014] Preferably, in step S4, the types of non-normalized interference include: Semantic ambiguity interference, the identification criteria of which include polysemous ambiguity and incoherent reference in the text. The logical coherence is below a preset threshold; Formatting error interference, the identification criteria for which the formatting error interference includes the layout features not matching the preset standard answer template, or the font, font size, and indentation format of the text being inconsistent; Logical confusion interference, the identification criteria for which include the problem-solving logical characteristics indicating missing reasoning steps, misuse of conditions, or contradictions.

[0015] Preferably, in step S5, the formulation and execution of the corresponding processing strategy specifically includes: for semantically ambiguous interference, retrieving relevant contextual information from a preset knowledge base and generating at least one semantically clear alternative. Ask descriptive or clarifying questions and incorporate them into the analysis report; to address formatting errors, automatically adjust or restructure the layout of the assignment content according to the standard answer template. It also generates format modification annotations; and, based on the aforementioned domain knowledge base and reasoning rules, generates hints or points out missing reasoning steps to address logical inconsistencies and interference. Logical contradictions should be identified and incorporated into the analysis report.

[0016] Preferably, after step S5, the method further includes: S6. Output the analysis report through the user interface, and provide teacher review and confirmation functions as well as student viewing functions. With feedback functionality.

[0017] Secondly, a system for handling non-standardized operational interference includes: The system includes a data acquisition module for receiving and preprocessing the operational data; a perception processing module for executing the processing logic of the perception model to extract the perception features; a cognition processing module for executing the processing logic of the cognition model to extract the cognitive features; and a fusion and recognition module for fusing the perception features and the cognitive features, and identifying the non-standardized interference class. The system includes a strategy execution and report generation module, used to execute processing strategies based on the identified interference types and generate the analysis. Report. Attached Figure Description

[0018] Figure 1 This is a flowchart of a method for handling non-standardized interference in operations according to the present invention. Detailed Implementation

[0019] The invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can implement it based on the description. The scope of protection of the invention is not limited to these specific embodiments.

[0020] This invention relates to a method for handling non-standardized interference in operations, such as... Figure 1 As shown, the method includes the following steps: S1. Receive job data to be processed, wherein the job data includes at least text data and / or image data.

[0021] Step S1 specifically includes: The system receives raw assignment data from a data source, which can be an image acquisition device, such as a high-speed scanner or camera, used to upload image data of paper assignments; it can also be an online teaching platform or learning management system, which directly obtains electronic document data or text data submitted by students through an application programming interface; or it can be a local or cloud storage device, used to read stored assignment files. The acquired raw job data is parsed to identify its data type and format. If the data is in image file format, it is identified as image data; if the data is plain text or a specific structured document format, the text data and possible layout information are parsed out. The parsed job data is standardized and preprocessed to form a unified internal data object for subsequent steps to call.

[0022] S2. Process the job data using a perception model to extract the perception features of the job data. Features include text content features and / or layout and graphic features extracted through image recognition.

[0023] Step S2 specifically includes: If the job data includes image data, the image data is preprocessed, including grayscale conversion, etc. At least one of binarization, denoising, and tilt correction is performed; based on optical character recognition (OCR) technology, text recognition is performed on the preprocessed image data to extract text content and its... Location information is used as a feature of the text content; based on image recognition technology, the preprocessed image data is analyzed to locate the areas containing the questions and the answers. Multiple structured areas, including the chart area, are extracted, and layout features describing the location, size, and relative relationship of each area are extracted. Non-textual visual elements are identified and extracted from the preprocessed image data to obtain graphic features that describe graphic categories, geometric properties, or chart structures. If the job data contains text data, the text data is encoded and standardized preprocessed, and the preprocessed data is then... The processed text data serves as the text content feature.

[0024] As the front-end of the entire processing flow, the perceptual model plays a crucial role in the initial processing and feature extraction of raw data such as images and text in the job. In practical applications, Optical Character Recognition (OCR) technology is one of the core technologies for perceptual models to process text data. When faced with a scanned job document image, OCR technology first performs grayscale processing, converting a color image into a grayscale image. This simplifies the image's complexity and facilitates subsequent processing. Next, through binarization, the grayscale image is further converted into a binary image, making the text in the image black and the background white, which helps to highlight the text information and facilitates subsequent text extraction and recognition. Afterward, denoising processing is performed to remove noise, stains, and other clutter from the image, improving the clarity of the text. Simultaneously, tilt correction is performed to adjust the image orientation, ensuring that the text is horizontally aligned and avoiding recognition errors. After completing these preprocessing steps, OCR technology uses its internal character recognition algorithm, based on deep learning models such as Convolutional Neural Networks (CNNs), to recognize the text in the image character by character, converting the text image into an editable text format.

[0025] Image recognition technology plays a crucial role in processing assignments containing graphics and charts. It uses edge detection algorithms to determine the contours of objects in an image, and then employs contour analysis algorithms to accurately locate the answer area. For example, in a math assignment, for geometry-related questions, image recognition technology can quickly identify the contours of triangles, circles, and other shapes, and determine their positions within the image. When locating the answer area, this technology also incorporates image feature information, such as color and texture, to further improve accuracy. Furthermore, to handle the complexity of different assignments, image recognition technology employs multi-scale analysis methods. By processing images at different scales, it can better detect answer areas and graphics of varying sizes, improving the algorithm's adaptability and robustness.

[0026] S3. Based on the perceptual features of the task data, process the task data through a cognitive model to extract the cognitive features of the task data. The cognitive features include semantic understanding features extracted based on natural language processing (NLP) and / or problem-solving logic features extracted based on logical reasoning.

[0027] Step S3 specifically includes: Based on the text content features, a natural language processing (NLP) model is used to perform lexical, syntactic, and deep semantic analysis, extract the semantic representation vectors and key semantic relationships of the text, and generate the semantic understanding features. Based on the text content features, layout features, and graphic features, the task content is transformed into a structured representation. A preset domain knowledge base and reasoning rules are invoked to perform logical verification and problem-solving path analysis, generating the problem-solving logic features that describe the problem-solving approach.

[0028] The cognitive model builds upon the features extracted by the perceptual model, conducting deeper semantic understanding and logical reasoning analyses to achieve a comprehensive understanding and judgment of the assignment content. In judging the correctness of answers, the cognitive model utilizes semantic understanding techniques from natural language processing. When faced with a Chinese reading comprehension question, the model first performs lexical analysis on the question text and the student's answer text, segmenting the text into individual words or phrases and labeling their parts of speech. Next, it performs syntactic analysis, constructing a grammatical structure tree of the sentences to understand their grammatical relationships. Based on this, through semantic analysis, utilizing a semantic knowledge base and deep learning models, it understands the semantic meaning of words and sentences in the text, grasping the text's theme and core viewpoints. Finally, it matches and compares the semantics of the student's answer with the semantics required by the question to determine whether the answer accurately addresses the question and covers the key points.

[0029] To identify problem-solving strategies, taking proof problems in math homework as an example, the cognitive model employs logical reasoning techniques. The model first identifies the given conditions and the conclusion to be proven in the problem, transforming this information into logical expressions. Then, based on existing mathematical knowledge and reasoning rules, it constructs a reasoning path. The cognitive model searches its knowledge base for relevant theorems, formulas, and reasoning patterns, attempting to deduce the conclusion step-by-step from the given conditions. During the reasoning process, the model continuously adjusts its reasoning strategy based on the results, selecting the optimal reasoning path. If contradictions are encountered or the reasoning cannot continue, the model reverts to previous steps, re-selects a reasoning direction, until a reasonable solution is found or the student's answer is determined to be incorrect.

[0030] S4. Integrate the perceptual features and the cognitive features to identify the types of non-standardized interference in the work data.

[0031] Step S4 specifically includes: The fusion of the perceptual features and the cognitive features specifically includes at least one of the following methods: data layer fusion, feature layer fusion, and decision layer fusion; wherein, Data layer fusion involves associating and aligning the original task image data with the text content features to form multimodal fusion data; feature layer fusion involves fusing at least two of the text content features, layout features, graphic features, semantic understanding features, and problem-solving logic features through a neural network to generate a joint feature representation; decision layer fusion involves obtaining a first judgment about the surface morphology of the task based on the perceptual features, obtaining a second judgment about the quality of the task content based on the cognitive features, and making a comprehensive decision based on the first and second judgments to obtain the identification result of the non-standardized interference type.

[0032] The invention employs a combination of data-level fusion, feature-level fusion, and decision-level fusion to fully leverage the advantages of perceptual and cognitive models and enhance the ability to handle non-standardized interference. In data-level fusion, the image and text data are fused at the initial stage of inputting the task into the system. Scanned task image data and text data obtained through OCR recognition are integrated to form a comprehensive data body containing both image and text information. The advantage of this is that subsequent processing can utilize the raw data information from both images and text, providing the model with more comprehensive input and enhancing its overall perception of the task content. When processing a physics task containing both graphics and text descriptions, data-level fusion allows the model to comprehensively consider image information such as the shape and size of the graphics, as well as textual information such as the physical principles and conditions described in the text, from the outset, avoiding the loss of crucial information that might occur when processing images or text separately.

[0033] Feature-level fusion involves fusing the features extracted from the data by the perceptual and cognitive models respectively. Image and text features extracted by the perceptual model, such as text features from OCR recognition and answer region features from image recognition, are combined with features extracted by the cognitive model based on semantic understanding and logical reasoning, such as semantic features of the question and logical features of the problem-solving approach. These features are then concatenated or weighted and fused using specific fusion algorithms. Through feature-level fusion, low-level features from the perceptual stage can be combined with high-level features from the cognitive stage, providing richer and more representative feature vectors for subsequent analysis and decision-making, thus improving the model's analytical capabilities and accuracy.

[0034] At the decision-making level, the perceptual and cognitive models make decisions based on their respective analysis results, and then these decisions are merged. The perceptual model analyzes the features of images and text to judge some surface information in the assignment, such as the existence of the answer area and the clarity of the text; the cognitive model, based on semantic understanding and logical reasoning, makes decisions on the correctness and reasonableness of the assignment content. The decisions of the two models are comprehensively considered, and a final decision result is obtained through voting mechanisms, weighted averaging, and other methods. When judging whether the solution to a math word problem is correct, the perceptual model judges whether the student's handwriting is clear, and the cognitive model judges whether the problem-solving approach and calculation process are correct. Through decision-making fusion, the results of the two judgments are combined to arrive at the final evaluation of the problem.

[0035] S5. Based on the identified non-standardized interference types, formulate and implement corresponding processing strategies to generate an analysis report that includes interference correction suggestions or content clarification.

[0036] Step S5 specifically includes: addressing semantically ambiguous interference by retrieving relevant contextual information from a pre-defined knowledge base and generating at least one semantically clear alternative. Ask descriptive or clarifying questions and incorporate them into the analysis report; to address formatting errors, automatically adjust or restructure the layout of the assignment content according to the standard answer template. It also generates format modification annotations; and, based on the aforementioned domain knowledge base and reasoning rules, generates hints or points out missing reasoning steps to address logical inconsistencies and interference. Logical contradictions should be identified and incorporated into the analysis report.

[0037] By fusing perceptual and cognitive models, the system can accurately identify various non-standardized interferences. For semantically ambiguous interference, the system first uses OCR technology in the perceptual model to extract the text information from the task and convert it into a computer-processable text format. Then, the cognitive model uses natural language processing techniques to analyze these texts in depth. Based on word vector models, such as Word2Vec or GloVe, each word is mapped to a vector in a low-dimensional vector space, and the semantic relationship between words is determined by calculating the similarity between vectors. For sentence semantic understanding, deep learning models, such as recurrent neural networks (RNNs) and their variants Long Short-Term Memory (LSTM) and Gated Recurrent Units (GRUs), are used to model the word sequence in the sentence and capture its semantic features. Through these techniques, the system can identify semantically ambiguous parts of the text, such as the ambiguity of polysemous words and semantic ambiguity caused by vague sentence structure. When processing the sentence "He walked for an hour," the system analyzes the multiple possible semantics of the word "walk," combining contextual information to determine whether it means "leave" or "walk."

[0038] To address formatting errors, the cognitive model uses image recognition technology to perform a preliminary analysis of the overall layout and format of the assignment. Utilizing edge detection and contour analysis algorithms, it identifies the position and shape of elements such as text areas, chart areas, and answer boxes, determining whether they conform to standard formats. For text formatting, the cognitive model checks whether the font, font size, line spacing, and indentation are consistent and standardized. In a math assignment, the system checks whether the solution steps are written according to the prescribed format, such as whether there are clear step numbers and whether correct mathematical symbols and formulas are used. If it finds that a formula in a step does not use standardized mathematical symbols, or that the solution steps are not numbered as required, the system will identify this as a formatting error.

[0039] To address different types of interference, this invention has developed corresponding processing strategies. For semantically ambiguous interference, the system utilizes a vast knowledge base and semantic analysis techniques to eliminate ambiguity. The knowledge base stores rich vocabulary, grammar, semantics, and domain knowledge. When encountering semantically ambiguous situations, the system searches for relevant information in the knowledge base and combines it with semantic analysis results to determine the most reasonable interpretation. When analyzing the sentence "an apple fell to the ground," if the word "apple" is ambiguous in a specific context (e.g., it could refer to the fruit apple or Apple Inc. products), the system will determine, based on the context and semantic information in the knowledge base, that "apple" most likely refers to the fruit, thus eliminating ambiguity. Simultaneously, semantic similarity calculation and semantic reasoning techniques are used to supplement and refine ambiguous semantics, making the expression clearer and more accurate.

[0040] When faced with formatting errors, the system will correct the assignment according to a pre-defined standard format. For text formatting issues, the system will automatically adjust the font, font size, line spacing, etc., to ensure compliance with standards. If inconsistent fonts are found in a section of text, the system will uniformly adjust them to the specified font. For non-standard answer formats, the system will reorganize the answer content according to the question type and requirements to conform to the standard answer format. When processing mathematical proof problems, if the student's answer steps are disorganized, the system will reorganize and format the known conditions, reasoning process, and conclusion according to the standard format for proof problems, making the solution process clearer and more organized.

[0041] S6. Output the analysis report through the user interface, and provide teacher review and confirmation functions as well as student viewing functions. With feedback functionality.

[0042] This invention also relates to a system for handling non-standardized operational interference, comprising: The system includes a data acquisition module for receiving and preprocessing the operational data; a perception processing module for executing the processing logic of the perception model to extract the perception features; a cognition processing module for executing the processing logic of the cognition model to extract the cognitive features; and a fusion and recognition module for fusing the perception features and the cognitive features, and identifying the non-standardized interference class. The system includes a strategy execution and report generation module, used to execute processing strategies based on the identified interference types and generate the analysis. Report.

[0043] The following example, a Chinese reading comprehension assignment, demonstrates the practical operation and effect of the method described in this invention. Suppose the question is, "Please analyze the meaning of the underlined sentence in the text." A student's answer is, "This sentence seems to express an emotion, but I can't quite put my finger on it," which clearly indicates semantic ambiguity. The system first extracts textual information using a perceptual model, and then analyzes the answer using a cognitive model. Utilizing word vector models and deep learning models, the system finds that expressions like "seems" and "can't quite put my finger on it" in the answer indicate semantic ambiguity. Next, the system searches the knowledge base for semantic information related to the sentence, and, combined with contextual analysis, determines that the student may have understood the emotion implied in the sentence, but their expression is unclear. The system further analyzes the keywords and semantic relationships in the sentence, and uses semantic reasoning technology to generate a clearer explanation, such as "This sentence expresses the author's [specific emotion] through [specific vocabulary or expression techniques]," and provides this as feedback to the student to help them clarify their thinking and improve their answer.

[0044] Taking a math homework assignment with formatting errors as an example, suppose a student, when solving a word problem, did not list the required step-by-step calculations but instead directly provided the final answer. The system, through a perception model, identifies that the answer format does not conform to the standard. Then, it re-organizes the solution process according to the standard answer format for math word problems. The system prompts the student to supplement the calculation process and thought process for each step, dividing the solution process into parts such as analysis of known conditions, application of formulas, calculation steps, and the final answer. This ensures that the homework format is standardized, the problem-solving thought process is clear, and it is convenient for teachers to grade and for students to self-check. These examples demonstrate that the processing method of this invention can effectively identify and handle non-standardized interference in student homework, improve the accuracy and efficiency of homework grading, and provide students with more targeted guidance.

Claims

1. A method for handling non-standardized interference in operations, characterized in that, The method includes the following steps: S1. Receive job data to be processed, wherein the job data includes at least text data and / or image data; S2. Process the job data using a perception model to extract the perception features of the job data. Features include text content features and / or layout and graphic features extracted through image recognition; S3. Based on the perceptual features of the homework data, the homework data is processed through a cognitive model to extract the cognitive features of the homework data. The cognitive features include semantic understanding features extracted based on natural language processing (NLP) and / or problem-solving logic features extracted based on logical reasoning. S4. By fusing the perceptual features and the cognitive features, the non-standardized interference types present in the job data are identified. S5. Based on the identified non-standardized interference types, formulate and implement corresponding processing strategies to generate an analysis report that includes interference correction suggestions or content clarification.

2. The method for handling non-standardized interference in operations according to claim 1, characterized in that, In step S2, the processing of the job data through the perception model specifically includes: If the job data includes image data, the image data is preprocessed, and the preprocessing includes at least one of grayscale conversion, binarization, denoising, and tilt correction. Based on optical character recognition (OCR) technology, text recognition is performed on preprocessed image data to extract text content and its location information as text content features. Based on image recognition technology, the preprocessed image data is analyzed to locate multiple structured regions, including question area, answer area, and chart area, and layout features describing the location, size and relative relationship of each region are extracted. Non-textual visual elements are identified and extracted from the preprocessed image data to obtain graphic features that describe graphic categories, geometric properties, or chart structures. If the job data contains text data, the text data is encoded and standardized preprocessed, and the preprocessed text data is used as the text content feature.

3. The method for handling non-standardized interference in operations according to claim 2, characterized in that, In step S3, the processing of the task data using a cognitive model specifically includes: Based on the text content features, a natural language processing (NLP) model is used to perform lexical, syntactic, and deep semantic analysis, extract the semantic representation vectors and key semantic relationships of the text, and generate the semantic understanding features. Based on the text content features, layout features, and graphic features, the task content is transformed into a structured representation. A preset domain knowledge base and reasoning rules are invoked to perform logical verification and problem-solving path analysis, generating the problem-solving logic features that describe the problem-solving approach.

4. The method for handling non-standardized interference in operations according to claim 3, characterized in that, In step S4, the fusion of the perceptual features and the cognitive features specifically includes at least one of the following methods: data layer fusion, feature layer fusion, and decision layer fusion; wherein, Data layer fusion involves associating and aligning the original task image data with the text content features to form multimodal fusion data; feature layer fusion involves fusing at least two of the text content features, layout features, graphic features, semantic understanding features, and problem-solving logic features through a neural network to generate a joint feature representation; decision layer fusion involves obtaining a first judgment about the surface morphology of the task based on the perceptual features, obtaining a second judgment about the quality of the task content based on the cognitive features, and making a comprehensive decision based on the first and second judgments to obtain the identification result of the non-standardized interference type.

5. The method for handling non-standardized interference in operations according to claim 4, characterized in that, In step S4, the types of non-normalized interference include: Semantic ambiguity interference, the identification criteria of which include the presence of polysemous words in the text. Ambiguity, unclear reference, or logical coherence below a preset threshold; Formatting error interference, the identification criteria for which the formatting error interference includes the layout features not matching the preset standard answer template, or the font, font size, and indentation format of the text being inconsistent; Logical confusion interference, the identification criteria for which include the problem-solving logical characteristics indicating missing reasoning steps, misuse of conditions, or contradictions.

6. The method for handling non-standardized interference in operations according to claim 5, characterized in that, In step S5, the formulation and execution of the corresponding processing strategy specifically includes: for semantic ambiguity interference, retrieving relevant contextual information from a preset knowledge base and generating at least one semantic... Clear alternative statements or clarifying questions should be incorporated into the analysis report. To address formatting errors, the layout of the assignment content is automatically adjusted or restructured based on the standard answer template, and formatting modification annotations are generated. To address logical inconsistencies and interference, based on the aforementioned domain knowledge base and reasoning rules, missing reasoning step hints or logical contradictions are generated and incorporated into the analysis report.

7. A method for handling non-standardized interference in operations according to claim 1 or claim 6, Its features are, Following step S5, the following is also included: S6. Output the analysis report through the user interface and provide a teacher review and confirmation function. And student viewing and feedback functions.

8. A system for handling non-standardized operational interference, used to implement claims 1 to 7. A method for handling non-standardized interference in operations as described in any one of the above, characterized in that, include: The data acquisition module is used to receive and preprocess the operation data; A perception processing module is used to execute the processing logic of the perception model to extract the perception features; A cognitive processing module is used to execute the processing logic of the cognitive model to extract the cognitive features; The fusion and recognition module is used to fuse the perceptual features and the cognitive features, and to identify the non-standard features. Standardize interference types; the policy execution and report generation module is used to execute processing policies based on the identified interference types and generate reports. The analysis report is as described above.

Images