Data processing method, system and electronic device based on intelligent correction

Abstract

This invention discloses a data processing method, system, and electronic device based on intelligent grading, relating to the field of educational informatization technology. The method includes: parsing standard answers using natural language processing, extracting standard knowledge point nodes and connecting edges, assigning logical priorities and calculating node weights, and constructing a standard cognitive path graph; parsing student answers, mapping student knowledge point nodes through semantic matching, and constructing a student cognitive path graph; comparing the two graphs, generating score labels based on node matching states and order offsets; detecting and completing missing nodes and broken paths, constructing a residual score structure graph; calculating structured scores, and generating a report with explanatory feedback. The system contains five corresponding modules, and the device includes a memory and a processor. This invention improves the accuracy and interpretability of grading, and the output report is adapted to student feedback and teacher review, making it suitable for automated grading in educational informatization.

Description

Technical Field

[0001] This invention relates to the field of educational informatization technology, and in particular to data processing methods, systems and electronic devices based on intelligent grading. Background Technology

[0002] With the deepening development of educational informatization, intelligent grading technology for text-based assignments has gradually become an important direction for the integration of natural language processing and intelligent education. Current mainstream methods mostly rely on natural language processing tools to perform grammatical analysis, keyword extraction, and semantic similarity calculation on student responses, and then combine this with standard answer templates to execute grading logic, achieving automatic scoring and basic feedback output. Especially in scenarios with many subjective questions, such as Chinese, politics, and history, intelligent grading systems are gradually replacing some manual operations, significantly improving grading efficiency. Matching mechanisms based on semantic vector calculation and scoring methods assisted by dependency parsing have become standard technical approaches in this field. These systems can provide unified and traceable preliminary scoring references for large-scale assignments, playing a positive role in educational equity and supporting teaching research.

[0003] However, existing intelligent grading technologies still have two limitations when dealing with complex cognitive expressions: First, these methods usually focus on matching the semantic similarity between the answer and the standard, but fail to construct a structured representation of the student's problem-solving process. Therefore, they are limited in capturing the sequential relationship, logical connection, and order deviation of knowledge points, and are unable to effectively reveal the deep cognitive biases in students' knowledge mastery. Second, existing feedback mechanisms often remain at the level of score or keyword prompts, lacking a structured and explainable expression of redundancy, omissions, and logical breaks in students' answers. As a result, feedback is not enough to fully support students' targeted corrections and teachers' efficient review. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a data processing method based on intelligent grading to solve the problem of being unable to quantify students' cognitive structure biases and generate structured explanatory feedback.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a data processing method based on intelligent grading, which includes: constructing a standard cognitive path graph based on standard answer text; parsing the text using natural language processing tools; constructing a set of nodes and a set of connecting edges for standard knowledge points; assigning logical priorities; calculating the weights of standard knowledge point nodes; and generating a standard cognitive path graph.

[0008] Using the standard cognitive path map as a reference, the student's answer text is analyzed and semantic matching is performed to map the student's knowledge point nodes to the standard knowledge point nodes, thus constructing the student's cognitive path map.

[0009] Based on the standard cognitive path map and the student cognitive path map, the matching status and order offset of student knowledge point nodes are compared and analyzed to generate a set of student knowledge point node scoring annotation information.

[0010] Verify the standard cognitive path graph, the student cognitive path graph, and the set of student knowledge point scoring annotation information; annotate the node status and insert missing nodes; perform path mapping and break detection; enumerate and select candidate edges at break points; and construct the graph residual scoring structure graph.

[0011] Based on the residual scoring structure graph, combined with the student's knowledge point node scoring annotation information, the structured score is calculated, the scoring result structure and explanatory feedback structure are generated, and the final scoring report is output.

[0012] As a preferred embodiment of the intelligent grading-based data processing method of the present invention, the specific steps for constructing the node set and connection edge set of standard knowledge points are as follows:

[0013] Perform semantic structure parsing to generate a standard set of sentences and a standard dependency syntax tree;

[0014] Phrases with grammatical rationality and subject knowledge carrying capacity are extracted from the standard dependency syntax tree and encapsulated into standard knowledge point nodes, which are then mapped to a set of standard knowledge point nodes.

[0015] Extract the original sentence number to which the standard knowledge point node belongs from the standard knowledge point node set, and construct the connection edge set of the standard knowledge point by combining the relative position of the standard knowledge point node in the standard answer text and the semantic logical relationship of the dependency syntax tree.

[0016] As a preferred embodiment of the data processing method based on intelligent grading described in this invention, the generation of the standard cognitive path graph refers to determining the connection edge type of the standard knowledge points and assigning the logical priority of the connection edge of the standard knowledge points, calculating the node weight of the standard knowledge points, and constructing the standard cognitive path graph.

[0017] As a preferred embodiment of the data processing method based on intelligent grading described in this invention, the steps of parsing student answers and performing semantic matching, mapping student knowledge point nodes to standard knowledge point nodes, and constructing a student cognitive path graph using a standard cognitive path graph as a reference, are as follows:

[0018] Perform semantic structure parsing of student response texts to generate a set of student sentences and a student dependency syntax tree. Extract phrases to construct and filter preliminary semantic units, and generate an initial set of student candidate knowledge point nodes. Perform preliminary matching of preliminary semantic units with a standard knowledge point vocabulary database to generate preliminary semantic mapping relationships and associate them with student candidate knowledge point nodes, and generate a set of student candidate knowledge point nodes.

[0019] Analyze the features of student candidate knowledge points and combine them to generate semantic representations of student candidate knowledge point nodes;

[0020] Calculate the semantic similarity score, compare the semantic similarity score with the set similarity threshold, and generate a set of mapping results between student candidate knowledge point nodes and standard knowledge point nodes, as well as an updated set of student knowledge point nodes;

[0021] Construct logical dependencies between student knowledge point nodes, generate a set of connection edges for student knowledge points and mark the connection edge types of student knowledge points, and construct a student cognitive path graph.

[0022] As a preferred embodiment of the data processing method based on intelligent grading described in this invention, the specific steps for comparing and analyzing the matching status and order offset of student knowledge point nodes to generate a set of student knowledge point node scoring annotation information are as follows:

[0023] Based on the semantic mapping results of student knowledge point nodes, the correspondence between student knowledge point nodes and standard knowledge point nodes is confirmed, the matching status is marked, and the mapping relationship between student knowledge point nodes and standard knowledge point nodes is generated.

[0024] The order offset is calculated by extracting the answer order index and the corresponding standard knowledge point node order number from the successfully matched student knowledge point nodes.

[0025] Map the sequential offset of each successfully matched student knowledge point node to a sequential level label;

[0026] Based on the sequential level labels, combined with the mapping relationship between student knowledge point nodes and standard knowledge point nodes, and the rating level annotations, a rating triplet for each student knowledge point node is generated. By integrating the rating triplets of all student knowledge point nodes, a set of student knowledge point node rating annotation information is generated.

[0027] As a preferred embodiment of the data processing method based on intelligent grading described in this invention, the following steps are taken: The verification standard cognitive path graph, the student cognitive path graph, and the student knowledge point scoring annotation information set are used to annotate node states and insert missing nodes.

[0028] Perform consistency checks on the standard cognitive path map, the student cognitive path map, and the set of student knowledge point node scoring annotation information, generate independent sessions, create event queues and log buffers, and set an upper bound for the maximum path length.

[0029] Each student's knowledge point node is labeled with its color and status. For missing nodes, virtual nodes are inserted and missing events are reported. The maximum path length is tightened, and a student cognitive path map with missing nodes and a list of colored nodes are generated.

[0030] As a preferred embodiment of the intelligent grading-based data processing method of the present invention, the specific steps of executing path mapping and fracture detection, enumerating and selecting candidate edges at fracture points, and constructing a graph residual scoring structure graph are as follows:

[0031] Traverse the student cognitive path graph after inserting the missing node, generate the student answer path sequence, register adjacent pairs, determine the attributes of the student answer path, and generate a list of broken edge locations.

[0032] Within the upper bound of the maximum path length, candidate path sequences are generated by enumerating. When a candidate path sequence cannot be generated, an event of insufficient candidates is constructed and the maximum path length is widened, and candidate paths are enumerated again.

[0033] The path selector prioritizes the path with the shortest number of steps, combines logical priority and standard order to generate a candidate priority table, performs edge patching or preserves breaks, updates the student's answer path and attributes, and outputs the graph residual scoring structure graph.

[0034] As a preferred embodiment of the intelligent grading-based data processing method described in this invention, the steps include: calculating structured scores based on the graph residual scoring structure graph, combining student knowledge point node scoring annotation information, generating a scoring result structure and an explanatory feedback structure, and outputting a final scoring report.

[0035] Based on the graph residual scoring structure graph, combined with the student knowledge point node scoring annotation information, a structured score is calculated using a fixed score rule and a structured feedback group is generated.

[0036] Based on structured scores and structured feedback groups, a scoring result structure and an explanatory feedback structure are generated;

[0037] Based on the scoring result structure and the explanatory feedback structure, a final scoring report is generated and used for student feedback and teacher review scenarios.

[0038] Secondly, the present invention provides a data processing system based on intelligent grading, comprising,

[0039] The standard graph construction module is used to parse the standard answer text, generate a set of standard knowledge point nodes and a set of connecting edges, assign logical priorities, calculate node weights, and generate a standard cognitive path graph.

[0040] The student knowledge graph generation module is used to parse student answer text, generate candidate nodes for student knowledge points, and perform semantic mapping with standard knowledge point nodes to construct a student cognitive path graph.

[0041] The scoring and annotation module is used to compare the standard cognitive path map and the student cognitive path map, analyze the matching status and sequence offset of student knowledge point nodes, and generate a set of scoring and annotation information for student knowledge point nodes.

[0042] The residual graph construction module is used to verify the standard cognitive path graph, the student cognitive path graph, and the student knowledge point scoring annotation information set, annotate the node status and insert missing nodes, perform path mapping and break detection, enumerate and select candidate edges at the break points, and construct the graph residual scoring structure graph.

[0043] The scoring and feedback module, based on the residual scoring structure graph and combined with student knowledge point node scoring annotation information, calculates structured scores, generates scoring result structures and explanatory feedback structures, and outputs the final scoring report. Thirdly, this invention provides a computer device, including a memory and a processor. The memory stores a computer program, wherein: when executed by the processor, the computer program implements any step of the data processing method based on intelligent grading as described in the first aspect of this invention.

[0044] The beneficial effects of this invention are as follows: By constructing a graph-based residual scoring structure graph, it automatically identifies and completes missing nodes, detects broken paths and inserts candidate edges based on the student's cognitive path graph. Simultaneously, it annotates node states and connection attributes, thus comprehensively depicting potential omissions, disordered sequences, and logical breaks in students' answers, enabling the grading process to possess structured, visual, and hierarchical expressive capabilities. By generating an explanatory feedback structure, it combines scoring results with path residual information to form a structured feedback group containing node order, connection status, and scoring annotations. This outputs actionable reports for both students and teachers, guiding students in targeted corrections and facilitating quick review by teachers. Ultimately, it achieves the beneficial effects of transparent scoring, quantifiable cognitive biases, and enhanced feedback direction in the grading process, significantly improving the application value of intelligent grading systems in teaching assistance and educational evaluation. Attached Figure Description

[0045] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1 This is a flowchart of a data processing method based on intelligent grading.

[0047] Figure 2 This is a flowchart for constructing a standard cognitive path map for data processing methods based on intelligent grading.

[0048] Figure 3 This is a flowchart for constructing a student cognitive path map based on a data processing method for intelligent grading.

[0049] Figure 4 This is a flowchart for generating a scoring report based on a data processing method using intelligent grading. Detailed Implementation

[0050] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0051] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0052] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the invention. The phrase "in one embodiment" appearing in different places throughout this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0053] Reference Figures 1-4 This is one embodiment of the present invention, which provides a data processing method based on intelligent grading, including the following steps:

[0054] S1. Construct a standard cognitive path graph based on the standard answer text. Use natural language processing tools to parse the text, construct a set of nodes and a set of connecting edges for standard knowledge points, assign logical priorities, calculate the weights of standard knowledge point nodes, and generate a standard cognitive path graph.

[0055] A standard cognitive path graph is constructed as the basic reference structure for subsequent comparison of students' cognitive path graphs and intelligent scoring. Taking the standard answer text as input, five sub-processes are executed in sequence: semantic structure parsing, standard knowledge point node extraction, logical relationship identification between nodes, path construction, and weight labeling. Finally, a standard cognitive path graph in a unified format is output.

[0056] Based on the standard answer text, semantic structure parsing is performed using natural language processing tools to generate a set of standard sentences and a standard dependency syntax tree.

[0057] Furthermore, semantic structure parsing is performed on the standard answer text using existing natural language processing tools.

[0058] Semantic structure parsing includes sentence boundary decomposition, word-level analysis, and dependency parsing, generating a standard sentence set and a standard dependency parsing tree.

[0059] Furthermore, the standard answer text is divided into sentence boundaries, and word-level analysis is performed on each sentence in the standard answer text to generate a set of standard sentences and a standard dependency syntax tree.

[0060] It should be noted that the existing natural language processing tools use Stanford CoreNLP.

[0061] Word-level analysis includes determining the grammatical category of each word through part-of-speech tagging, extracting key entities in the sentence through entity recognition, and analyzing the subject-verb, verb-object, and modification grammatical relationships between words in the sentence based on dependency syntax rules.

[0062] The standard dependency syntax tree presents the grammatical structure of a sentence in a tree-like format.

[0063] Phrases with grammatical rationality and subject knowledge content are extracted from the standard dependency syntax tree and encapsulated into standard knowledge point nodes, which are then mapped to a set of standard knowledge point nodes.

[0064] Furthermore, based on the standard sentence set and the standard dependency syntax tree, phrases with grammatical rationality and subject knowledge carrying capacity are extracted from the standard dependency syntax tree and mapped to a standard knowledge point node set.

[0065] Furthermore, based on predefined semantic templates and keyword triggering rules, phrases with grammatical rationality and subject knowledge carrying capacity are extracted from the standard dependency syntax tree and identified as standard knowledge point expression units. Each standard knowledge point expression unit is assigned a unique structured field and encapsulated into a standard knowledge point node. All standard knowledge point nodes are integrated to form a standard knowledge point node set.

[0066] It should be noted that phrases possess both grammatical validity and the capacity to carry subject-specific knowledge.

[0067] The dedicated structured fields include the standard knowledge point node number, standard knowledge point category, original sentence number, standard terminology phrase, and initial node weight value.

[0068] Standard knowledge point expression units are segments of subject-related content in the standard answer that are relevant to solving the problem, such as term definitions, formula statements, logical reasoning, and experimental conclusions.

[0069] Extract the original sentence number to which the standard knowledge point node belongs from the standard knowledge point node set, and construct the connection edge set of the standard knowledge point by combining the relative position of the standard knowledge point node in the standard answer text and the semantic logical relationship of the dependency syntax tree.

[0070] Furthermore, based on the order of appearance and semantic logic of standard knowledge points in the standard answer text, a set of connection edges for standard knowledge points is established.

[0071] Furthermore, extract each standard knowledge point node. The original sentence number is assigned, and the position of each standard knowledge point node within the sentence is marked to form the sequential numbering of the standard knowledge point nodes. For any two distinct nodes and ,like exist Previously, the connection edges of the candidate standard knowledge points were initially established, represented as follows:

[0072] ;

[0073] in, The connecting edges representing candidate standard knowledge points have potential sequential relationships.

[0074] Furthermore, by combining the semantic dependencies (such as subject-verb, verb-object, and modifiers) in the standard dependency syntax tree, it is determined whether the connection edges of the candidate standard knowledge points conform to the standard. yes Logical premise:

[0075] If node The semantic content directly references, interprets, or depends on the node. If the content is valid, it is considered a valid connection edge; if there is only a positional order but no semantic dependency, the connection edge of the candidate standard knowledge point is removed.

[0076] Furthermore, all valid standard knowledge points are aggregated to form a set of standard knowledge point connection edges.

[0077] The connection edge type of standard knowledge points is determined by a heuristic rule base, and the logical priority of the connection edge of standard knowledge points is assigned.

[0078] Furthermore, the connection types of standard knowledge points are determined through a heuristic rule base, including sequential edges, logical causal edges, and conditional dependency edges.

[0079] It should be noted that sequential edges represent the order of text, logical causal edges represent "therefore" or "so", and conditional dependency edges represent "if...then..." structures.

[0080] The heuristic rule base makes decisions based on conjunctions and dependency structures.

[0081] Furthermore, assign logical priorities to the connecting edges of each standard knowledge point. This is used to subsequently assess the degree of disorder in the order of students' answers.

[0082] The logical priority values ​​are (1, 2, 3), which represent "weak logical dependency", "medium-strength logical dependency" and "strong logical dependency" respectively.

[0083] It should be noted that weak logical dependency is characterized by a textual sequence relationship between knowledge points, without substantial semantic causal or structural dependencies. The sequence is determined by sentence number or statement position, lacking logical connectors.

[0084] Medium-strength logical dependency is characterized by the preceding knowledge point serving as the causal basis or premise for the following knowledge point, possessing a certain degree of logical explanatory power. The determination rule is the existence of logical causal connectors and causal or inductive relationships in dependency syntax.

[0085] Strong logical dependency is characterized by the fact that the expression of a subsequent knowledge point depends on the preceding knowledge point. The judgment rule is that there are obvious precondition words in the sentence and that the conditional subordination is reflected in the dependency syntax structure.

[0086] The weight of a standard knowledge point node is calculated based on the logical priority of the connection edges of the standard knowledge point and the relative position of the standard knowledge point node.

[0087] Furthermore, to ensure scoring sensitivity, each standard knowledge point node is assigned a weight based on its category and relative position, as follows:

[0088] ;

[0089] in, Representing standard knowledge point nodes The weight, The logical priority of the edge type to which the standard knowledge point node belongs is indicated, with values ​​of (1, 2, 3). This represents the weight adjustment coefficient, with a value range of (0,1). This indicates the sequential numbering of the standard knowledge point nodes.

[0090] It should be noted that the numerical range of the original sequential numbering may fluctuate significantly due to differences in sentence length and the density of knowledge point distribution. The sequential numbering of the standard knowledge point nodes needs to be normalized.

[0091] Furthermore, for all standard knowledge point nodes in the standard answer, find the minimum and maximum numbers among the sentence number and the original sequential number of the sentence to which the standard knowledge point node belongs, and calculate the sequential index of the standard knowledge point node, represented as:

[0092] ;

[0093] in, Representing standard knowledge point nodes Sequential index, This indicates the sentence number where the standard knowledge point node is located. This represents the original sequential number of a single standard knowledge point node within its sentence, obtained by following the natural reading order from the beginning to the end of the sentence. This represents the smallest numerical value among the original sequential numbers of all standard knowledge point nodes within the same sentence. This represents the number with the largest numerical value among the original sequential numbers of all standard knowledge point nodes within the same sentence.

[0094] It should be noted that the weight adjustment coefficient is used to adjust the contribution ratio of the logical priority of the edge to which the standard knowledge point belongs and the relative position of the standard knowledge point to the weight of the standard knowledge point node. Since different grading scenarios (such as subjects and question types) have different sensitivities to "logical priority" and "relative position", it needs to be determined in combination with the specific scenario.

[0095] Based on the standard knowledge point node set, the standard knowledge point connection edge set, the standard knowledge point connection edge logical priority, and the standard knowledge point node weight, a standard cognitive path graph is constructed by structural combination.

[0096] Furthermore, the standard knowledge point node set, the standard knowledge point connection edge set, the standard knowledge point node weight, and the standard knowledge point connection edge logical priority are included. By performing structured combination, a standard cognitive path map is generated, represented as follows:

[0097] ;

[0098] in, Represents the standard cognitive path map. This represents a set of standard knowledge point nodes. The set of connecting edges representing standard knowledge points.

[0099] It should be noted that the standard cognitive path graph is a set of directed weighted graph structures and is stored in the form of JSON graph description language or graph database structure; it will serve as a reference basis for comparing students' cognitive path graphs in subsequent steps.

[0100] S2. Using the standard cognitive path map as a reference, analyze the student's answer text and perform semantic matching to map the student's knowledge point nodes to the standard knowledge point nodes, thus constructing the student's cognitive path map.

[0101] Based on student responses, semantic structure parsing is performed using natural language processing tools to generate a set of student sentences and a student dependency syntax tree. Based on the set of student sentences and the student dependency syntax tree, phrases are extracted to construct preliminary semantic units. These preliminary semantic units are then filtered and merged to generate student candidate knowledge point nodes, which are then aggregated into an initial set of student candidate knowledge point nodes.

[0102] Furthermore, semantic structure parsing consistent with standard cognitive path maps is performed on student response texts, and natural language processing tools are invoked to perform sentence boundary delineation, word-level analysis, and dependency parsing to generate a set of student sentences and a student dependency parsing tree.

[0103] Furthermore, the results of dependency parsing extract grammatical relationships such as subject-predicate, verb-object, and modification between words in the sentence. Based on these grammatical relationships, the extracted semantically valuable phrases or words are grouped into preliminary semantic units. These preliminary semantic units are then selected and merged to generate candidate knowledge point nodes. The candidate knowledge point nodes are recorded and numbered sequentially, and then aggregated into an initial set of student candidate knowledge point nodes.

[0104] It should be noted that the initial semantic unit selection and merging process includes removing redundancy, merging similar core concept phrases, modifiers, and syntactic association markers.

[0105] The initial semantic units are matched with the standard knowledge point vocabulary to generate an initial semantic mapping relationship and associate it with student candidate knowledge point nodes, thus generating a set of student candidate knowledge point nodes.

[0106] Furthermore, based on the preliminary semantic units, a pre-defined standard knowledge point vocabulary is invoked, and the cosine similarity between the preliminary semantic units and each item in the standard knowledge point vocabulary is calculated using the word vector model in the natural language processing tool. At the same time, combined with syntactic structure alignment analysis, a preliminary matching score is generated.

[0107] It should be noted that the word vector model is based on Chinese word vector training using Skip-gram Negative Sampling (SGNS). The process involves collecting textbooks, standard answers, problem solutions, and anonymized student responses, completing source registration and compliance review; deduplication and noise reduction, standardizing encoding and simplified / traditional characters, standardizing number and time notation, and preserving the original form of subject terminology; sentence segmentation and word segmentation; adding subject terminology to a custom dictionary; calculating word frequencies, setting a minimum frequency threshold, and filtering noise words; establishing a whitelist for key terms; establishing a word-to-number mapping; extracting positive samples from the center word and context words using a sliding window; extracting several negative samples based on word frequency distribution; downsampling extremely high-frequency words; initializing two sets of vector matrices; and using skip-word sampling. The model employs a negative sampling objective function to maximize the score of positive samples and minimize the score of negative samples. It sets the batch size, learning rate, and number of rounds, updating parameters sentence by sentence. Training loss and logs are recorded, and snapshots are saved at intervals for resuming training at breakpoints. Internally, word pair similarity and synonymous nearest neighbor accuracy are used for evaluation, and validation is performed on small-sample semantic matching tasks when necessary. The model stops early when the metrics no longer improve. A vocabulary and vector files are exported, and the version and training configuration are recorded. Out-of-vocabulary word handling strategies are provided (e.g., splitting by character or word fragment and averaging the vectors, or using nearest neighbor substitution), and vector normalization is performed uniformly during the inference phase.

[0108] It should be noted that the standard knowledge point vocabulary includes authoritatively defined standard knowledge point terms, a set of synonyms, domain-specific vocabulary, and their corresponding semantic codes.

[0109] Syntactic alignment analysis includes comparing the parts of speech and modification relationships of preliminary semantic units with those of standard vocabulary in the standard knowledge point vocabulary database.

[0110] Furthermore, based on the preliminary matching scores, the top three standard vocabulary entries with the highest scores are selected as candidate matching items to form a preliminary semantic mapping relationship.

[0111] It should be noted that the preliminary semantic mapping relationship includes the unique identifier of the preliminary semantic unit, the matched standard word ID, the multi-dimensional similarity score, and the matching confidence.

[0112] The multi-dimensional similarity score includes literal similarity, semantic relevance, and domain suitability; the matching confidence score ranges from (0 to 1).

[0113] Furthermore, the generated preliminary semantic mapping relationship is associated with the student candidate knowledge point nodes. The preliminary semantic mapping relationship is established one-to-one with the student candidate knowledge point nodes through the unique identifier of the preliminary semantic unit, ensuring that each student candidate knowledge point node is associated with the preliminary semantic mapping relationship.

[0114] Furthermore, all student candidate knowledge point nodes with preliminary semantic mapping relationships are structurally integrated and sorted according to their order of appearance in the student's answer text to generate a set of student candidate knowledge point nodes.

[0115] It should be noted that the set of student candidate knowledge point nodes includes the student candidate knowledge point node number, text content, syntactic features, semantic tags, and a unique identifier of the semantic mapping relationship between the student candidate knowledge point nodes and the standard knowledge point nodes.

[0116] Furthermore, all preliminary semantic mapping relationships are categorized according to the student candidate knowledge point node numbers, and the matching path descriptions between student candidate knowledge point nodes and standard vocabulary are supplemented to form associated preliminary semantic mapping results.

[0117] The features of student candidate knowledge points are analyzed and combined to generate semantic representations of student candidate knowledge point nodes.

[0118] Furthermore, the set of student candidate knowledge point nodes and preliminary semantic mapping results are obtained. The features of student candidate knowledge points are analyzed by semantic matching and combined to generate the semantic representation of student candidate knowledge point nodes. The semantic representation of student candidate knowledge point nodes is compared with the semantic representation of standard knowledge point nodes one by one to generate a semantic similarity score.

[0119] Furthermore, semantic analysis is performed on the set of student candidate knowledge point nodes; through semantic analysis, lexical hierarchical features, syntactic hierarchical features, and semantic association features of student candidate knowledge point nodes are extracted.

[0120] The lexical and syntactic features of student candidate knowledge point nodes are weighted and fused to generate a preliminary semantic representation, ensuring that student candidate knowledge point nodes can be effectively compared with standard knowledge point nodes.

[0121] Furthermore, the semantic representation of the student candidate knowledge point node is compared one by one with the semantic representation of each standard knowledge point node in the standard knowledge point node set to generate a semantic similarity score between the student candidate knowledge point node and the standard knowledge point node.

[0122] It should be noted that the semantic similarity score reflects the degree of semantic proximity between student candidate knowledge point nodes and standard knowledge point nodes.

[0123] Each student candidate knowledge point node corresponds to a set of scoring data, including the standard knowledge point node ID, semantic similarity score, and a description of the scoring composition.

[0124] Furthermore, to make semantic similarity scoring more accurate, semantic similarity scores are often obtained based on a combination of multiple existing methods, including those utilizing the intersection ratio of word sets. Coefficients, obtained using term frequency and inverse document frequency The weight distribution and the semantic cosine similarity obtained based on the cosine angle of the vector space are expressed as:

[0125] ;

[0126] in, Indicates semantic similarity score, Representing a set of words Similarity Indicates the degree of matching based on term frequency. This represents a vectorized cosine similarity.

[0127] The semantic similarity score is compared with the set similarity threshold to generate a set of mapping results between student candidate knowledge point nodes and standard knowledge point nodes, as well as an updated set of student knowledge point nodes.

[0128] Furthermore, the semantic similarity score is compared with a set similarity threshold to determine whether the student candidate knowledge point node has a valid mapping relationship with a certain standard knowledge point node. If the semantic similarity score is greater than or equal to the set similarity threshold, the student candidate knowledge point node matches the corresponding standard knowledge point node, and the corresponding mapping relationship is recorded. If the semantic similarity score is lower than the set similarity threshold, the student candidate knowledge point node will not be mapped and will remain in an unmatched state.

[0129] It should be noted that the set similarity threshold was generated based on the automatic scoring results and the semantic matching accuracy distribution by collecting a certain number of standard answers and student answers as comparison samples during the initial training phase.

[0130] Furthermore, each comparison result generates a matching record, including the student candidate knowledge point node number, the mapped standard knowledge point node number, and the corresponding semantic similarity score; the matching records that meet the set semantic similarity threshold conditions are aggregated to generate a set of matching results between candidate knowledge point nodes and standard knowledge point nodes.

[0131] Furthermore, based on the matching result set between candidate knowledge point nodes and standard knowledge point nodes, and according to the mapping relationship between candidate knowledge point nodes and standard knowledge point nodes in each matching result, the student candidate knowledge point nodes are renumbered and assigned corresponding category labels, thereby generating an updated set of student knowledge point nodes.

[0132] It should be noted that the generated updated student knowledge point node category labels ensure that the updated student knowledge point nodes are semantically consistent with the standard knowledge point nodes, thereby ensuring that there will be no ambiguity in the semantic classification of student knowledge point nodes during the subsequent construction of the student cognitive path graph.

[0133] Based on the updated set of student knowledge point nodes, and combined with the sentence order in the student's answer text, the logical dependencies between student knowledge point nodes are constructed, a set of connection edges for student knowledge points is generated, and the connection edge types of student knowledge points are marked.

[0134] Based on the updated set of student knowledge point nodes and the order of sentences in the student's answer text, directed connections between student knowledge point nodes are constructed, forming a set of connection edges for student knowledge points.

[0135] Furthermore, the logical dependencies between knowledge point nodes in the student's answer text are identified based on the sentence order in the student's answer text; based on the identified logical dependencies between student knowledge point nodes, connection edges between student knowledge point nodes are constructed, and each connection edge is used to represent the semantic connection between two student knowledge point nodes.

[0136] Furthermore, the semantic structure of the students' answers is analyzed to generate a set of student sentences and a student dependency syntax structure; semantic units are extracted and filtered to generate candidate knowledge point nodes for students, and the order of each student knowledge point node is determined according to its order of appearance in the text.

[0137] According to the order of students' answers, adjacent student knowledge point nodes are paired up and registered as candidate start and end pairs to express the order of students' answers. For each pair of candidate start and end nodes, a heuristic rule is used to make a determination.

[0138] It should be noted that the criteria for determining heuristic rules include conjunctions and dependency syntax structures, and logical dependencies include sequential dependencies, causal dependencies, and conditional dependencies.

[0139] Among them, sequential dependency indicates only the order of textual relationships, without reflecting substantial semantic causality or structural subordination.

[0140] Causal dependency indicates that the statement of a subsequent knowledge point takes a previous knowledge point as a cause or premise, is accompanied by causal conjunctions, and reflects causal or inductive relationships in dependency relations.

[0141] Conditional dependency indicates that the validity of a subsequent knowledge point depends on a previous knowledge point. It often uses conditional conjunctions in sentences and presents conditional subordination in dependency relations.

[0142] A connection is considered valid when the semantic content of the endpoint node references, interprets, or depends on the content of the starting node; otherwise, the candidate connection is discarded if there is only a positional order but no semantic dependency.

[0143] For connections that are confirmed to be valid, they are classified into three levels: weak logical dependency, medium logical dependency, and strong logical dependency. Weak logical dependency corresponds to a simple sequential relationship, medium logical dependency corresponds to a causal or premise relationship, and strong logical dependency corresponds to a necessary condition relationship.

[0144] All valid connections are aggregated into a set of student connection edges, with each edge representing a semantic connection between two student knowledge point nodes; the student cognitive path graph is formed by combining the set of student knowledge point nodes.

[0145] It should be noted that the connection edges between students' knowledge points are marked with their types when they are generated; the edge type marking further enhances the interpretability of semantic relationships in the students' cognitive path graph, and can distinguish different logical chains in subsequent grading.

[0146] Based on the updated set of student knowledge points The set of edges connecting students' knowledge points Construct a student cognitive path map, represented as follows:

[0147] ;

[0148] S3. Based on the standard cognitive path map and the student cognitive path map, compare and analyze the matching status and order offset of student knowledge point nodes to generate a set of student knowledge point node rating annotation information.

[0149] Based on the semantic mapping results of student knowledge point nodes, the correspondence between student knowledge point nodes and standard knowledge point nodes is confirmed, the matching status is marked, and the mapping relationship between student knowledge point nodes and standard knowledge point nodes is generated.

[0150] Furthermore, based on the semantic mapping results of each student knowledge point node in the student cognitive path graph, the standard knowledge point node corresponding to the student knowledge point node in the standard cognitive path graph is identified. If a one-to-one correspondence exists, the matching status identifier is marked as "match successful". If the student knowledge point node does not have a corresponding standard knowledge point node, the matching status identifier is marked as "redundant node". If the standard knowledge point node has no matching student knowledge point node, the matching status identifier is marked as "missing node". The mapping relationship results between all student knowledge point nodes and standard knowledge point nodes are generated.

[0151] The order offset is calculated by extracting the answer sequence index and the corresponding sequence number of the standard knowledge point node from the successfully matched student knowledge point nodes.

[0152] Furthermore, based on the mapping results between all student knowledge point nodes and standard knowledge point nodes, student knowledge point nodes marked as successfully matched in the mapping results are... Extract the sequential index of student knowledge point nodes and their corresponding standard knowledge point nodes from the student's answer text. The sequential numbering in the standard cognitive path map is represented as follows:

[0153] ;

[0154] in, Representing student knowledge points Sequential index, Representing student knowledge points The sentence number in which it appears. Representing student knowledge points Sentence The phrase offset within.

[0155] It should be noted that, The value range is (0,1), reflecting the relative position of the student's knowledge point node within the sentence.

[0156] The function's range of values ​​is (1.0, ...). ),in The total number of sentences in the student's response text.

[0157] Furthermore, based on the sequential index of student knowledge point nodes in the student's answer text. and corresponding standard knowledge point nodes Sequential numbering in the standard cognitive path map Calculate the sequential offset of each successfully matched node. , is represented as:

[0158] ;

[0159] It should be noted that the order offset is a signed real number; a negative value indicates an early expression, a positive value indicates a late expression, and the larger the absolute value, the greater the order deviation.

[0160] The order offset level is determined by combining the order offset with two preset adjustable order consistency tolerance thresholds and a severe misorder threshold.

[0161] Furthermore, the sequence consistency tolerance threshold The value range is (0.3, 0.6); Severe out-of-order threshold. The value range is (1.0, 1.5).

[0162] It should be noted that the order consistency tolerance threshold and the out-of-order severity threshold can be automatically obtained based on the order deviation samples between the student's answer and the standard path. The natural clustering range of the order deviation is analyzed using the cumulative distribution function, and the threshold range is set to adapt to the semantic requirements of the task. The order consistency tolerance threshold and the out-of-order severity threshold can be dynamically adjusted according to the question type or historical data performance to improve the adaptability to the student's expression order tolerance.

[0163] Based on the order offset of each successfully matched student knowledge point node, combined with the preset adjustable order consistency tolerance threshold and misorder severity threshold, the order offset of each successfully matched student knowledge point node is mapped to an order level label.

[0164] Furthermore, the sequential offset of each successfully matched student's knowledge point node is... Mapped to three categories of sequential level labels , is represented as:

[0165] ;

[0166] Based on sequential level labels The mapping results between student knowledge point nodes and standard knowledge point nodes are used to generate rating level labels.

[0167] Furthermore, a rating level labeling is performed based on a combination rule of matching status and order level.

[0168] It should be noted that the matching status includes successful matching, redundancy, and missing; the order level includes consistent order, slight misorder, and severe misorder.

[0169] When a match is successful and the order level is "consistent order", the score is marked as full marks. When a match is successful and the order is "slightly out of order", the score is marked as partial marks. When a match is successful and the order is "severely out of order", redundant nodes, or missing nodes, the score is marked as zero marks.

[0170] Based on the sequential level labels, combined with the mapping relationship between student knowledge point nodes and standard knowledge point nodes, and the rating level annotations, a rating triplet for each student knowledge point node is generated. By integrating the rating triplets of all student knowledge point nodes, a set of student knowledge point node rating annotation information is generated.

[0171] Furthermore, the scoring information for each student's knowledge point node is defined as a triple, represented as:

[0172] ;

[0173] in, This represents a triplet representing the student's knowledge point node scoring information. It is a matching status identifier between student knowledge point nodes and standard knowledge point nodes, including successful matching, redundant nodes, and missing nodes. Indicates the rating level label, Indicates a sequence level label.

[0174] Furthermore, based on the triplet scoring information of student knowledge point nodes, the triplet scoring information corresponding to all student knowledge point nodes is structurally integrated to generate a set of student knowledge point node scoring annotation information. , is represented as:

[0175] ;

[0176] in, This indicates traversing the set of student knowledge point nodes. Each student knowledge point node Generate corresponding triplet score information for each node. , Representing the student's cognitive path map The set of nodes.

[0177] It should be noted that each record in the student knowledge point node rating annotation information set includes the student knowledge point node number, the matching standard knowledge point node number, the order offset, the order level, the matching status identifier, and the rating level.

[0178] S4. Verify the standard cognitive path graph, the student cognitive path graph, and the set of student knowledge point scoring annotation information. Mark the node status and insert missing nodes. Perform path mapping and break detection. Enumerate and select candidate edges at the break points. Construct the graph residual scoring structure graph.

[0179] Perform consistency checks on the standard cognitive path map, the student cognitive path map, and the set of student knowledge point node rating annotation information, generate independent sessions, create event queues and log buffers, and set an upper bound for the maximum path length.

[0180] Furthermore, the standard cognitive path graph, the student cognitive path graph, and the student knowledge point node scoring annotation information set are analyzed to perform consistency verification and generate an independent session for each answer. A session number is created and assigned to each independent session, and an event queue and log buffer are created based on the session number to serve as the record carrier for subsequent node insertion, path determination, and candidate edge completion.

[0181] Furthermore, a consistency check is performed on the standard cognitive path map, the student cognitive path map, and the set of student knowledge point node scoring annotation information.

[0182] It should be noted that consistency verification includes identity consistency, edge validity, and field completeness verification.

[0183] Furthermore, the verifier reads the student knowledge point node identifiers in the student cognitive path graph one by one and searches for records with the same name in the student knowledge point node rating annotation information set. The verifier outputs the matching nodes as a "list of annotable nodes" and outputs the nodes that exist as standard nodes but are missing student nodes as a "list of nodes to be inserted". The verifier writes the two lists and the corresponding reasons into the log buffer.

[0184] The verifier performs edge validity checks, reading each edge connecting the student's cognitive path graph and the standard cognitive path graph and checking whether the nodes at both ends exist in their respective node sets. The verifier marks or removes duplicate edges for detected dangling edges and duplicate edges, and outputs the cleaned edge set to the subsequent path determination unit. At the same time, the correction details are written to the log buffer for traceability.

[0185] The execution field completeness check involves the validator reading key fields from the scoring label information set, including matching status, order offset, order level, scoring level label, order consistency tolerance threshold, and out-of-order severity threshold. The validator performs default value completion for missing or out-of-bounds metadata and writes the "reason and source of completion" to the log. The validator does not modify the order level label. In subsequent processing stages, the order level label is used as the sole order criterion only during 'order determination'; scoring-related determinations are based on the scoring level label and are ultimately interpreted and effective by the scoring rules.

[0186] Furthermore, the event handling facilities are initialized. The event manager registers two types of event listeners, "missing" and "insufficient candidates," within the session scope, and outputs the triggering conditions and callback order of the two types of events to the candidate path enumerator and parameter manager so that the search boundary can be adjusted in real time when missing insertions or insufficient candidates occur.

[0187] Furthermore, the parameter manager is invoked to register runtime control variables. The parameter manager reads the session number and creates a new control variable table in the current session. The parameter manager sets the maximum path length to the default three steps and marks it as "flow control variable used only for upper bound of candidate edge search".

[0188] It should be noted that the parameter manager outputs the read handle of the control variable to the candidate path enumerator and keeps the write handle of the control variable only inside the parameter manager to ensure the atomicity and traceability of the modification.

[0189] It should be noted that the maximum path length is set and the default is three steps because candidate edge completion needs to enumerate paths from the break point to the break point in the set of connected edges of the standard cognitive path graph. Without setting an upper bound, there is a risk of scale expansion and semantic drift. The default value of three steps can cover most reasonable transitions in common problem types and achieve a trade-off between computational cost, edge completion quality and real-time performance. Then, in the session, the upper bound of candidate edge completion search is automatically tightened to two steps based on the "missing item found" event, and temporarily relaxed to four steps for individual breaks based on the "insufficient candidates" event, thereby keeping the balance between efficiency and quality on an event-driven track.

[0190] Furthermore, the "three types of verified input data", "session number and event queue handle", "log buffer handle", and "maximum path length read handle" are packaged together and output to the node labeling and missing insertion unit.

[0191] Using the verified set of student knowledge point node rating and annotation information as a reference, the color and status of each student knowledge point node are marked one by one. The results are compared with the verified standard cognitive path map. Virtual nodes are inserted for missing nodes and missing events are reported. The maximum path length is tightened, and a student cognitive path map with missing nodes and a list of colored nodes are generated.

[0192] Furthermore, based on the verified set of student knowledge point node rating annotation information, each student knowledge point node in the student cognitive path graph is traversed in the order of answering, and the matching status, order level label, order offset, and rating level annotation are read in sequence and marked with color and status.

[0193] During the same traversal, the standard cognitive path map is compared synchronously. For knowledge points that are present in the standard cognitive path map but missing in the student cognitive path map, virtual nodes are inserted at the corresponding positions in the student cognitive path map and marked in red. The status is recorded as missing insertion. The red virtual nodes are summarized to form a list of missing nodes. At the same time, the insertion position and the corresponding standard node number are recorded, and the missing event is reported when it first appears.

[0194] The labeled and inserted nodes are written to the colored student node list in real time and then populated back into the student cognitive path map.

[0195] It should be noted that when labeling node colors and statuses, a successful match with consistent order is marked in green, and the status is recorded as "matched and consistent order"; a successful match with slight or severe misorder is marked in yellow, and the status is recorded as "matched and slightly misordered" or "matched and severely misordered" respectively, with the corresponding order offset recorded; a node that cannot match the standard or is judged as student redundancy is marked in gray, and the status is recorded as "student redundancy" with a source explanation. Student nodes marked in gray are compiled to form a node redundancy list.

[0196] It should be noted that after all nodes have been processed, the node labeling and missing insertion unit generates a color-coded list of student nodes and a student cognitive path map with missing nodes inserted.

[0197] After receiving and verifying the "missing information found" event message, the parameter manager tightens the maximum path length from the default three steps to two steps, and outputs the "tightened two-step upper bound" to the candidate path enumerator. At the same time, it outputs the values ​​before and after tightening and the triggering reason to the log buffer for traceability.

[0198] Traverse the student cognitive path graph after inserting the missing node, generate the student answer path sequence, register adjacent pairs, determine the attributes of the student answer path, and generate a list of broken edge positions.

[0199] Furthermore, the student cognitive path graph after inserting the missing node is traversed, and combined with the list of colored student nodes, a sequence of student answer paths is generated. Adjacent pairs are registered, and the attributes of the student answer paths are determined by combining the standard connection edge set, the order offset of the two-end nodes, and the threshold, and a list of broken edge positions is generated.

[0200] Read the student cognitive path graph after inserting the missing node and the list of student nodes marked with color. Arrange the student answer path sequence according to the order of the student's answer and load the order consistency tolerance threshold and the misorder severity threshold. Traverse the list of student nodes marked with color according to the order of the student's answer. Form a pair of start and end nodes for every two adjacent student knowledge points and register them in the adjacent pair candidate table. Record the start point, end point and context position of each pair in the student cognitive path graph.

[0201] It should be noted that the adjacent candidate list transforms the sequential representation of students into the smallest computable decision unit, which not only accurately aligns with the standard graph, but also provides a low-cost and highly interpretable basis for subsequent break detection, candidate edge filling, and scoring and feedback.

[0202] Furthermore, each pair of start and end nodes in the adjacent candidate list is checked against the standard cognitive path graph. The connection edge set of the standard cognitive path graph is queried to see if there is a directed connection in the same direction as the start and end nodes, and the query result of "direct connection" or "no direct connection" is obtained. The order offset, color and missing flag of the start and end nodes in the scoring label information are read and input into the state determiner for three-state determination.

[0203] It should be noted that the state determiner reads the color and missing flag of the two-end nodes. When either end node is a red missing node, the state determiner directly determines the adjacent pair as a broken edge and writes the reason "broken due to missing" into the path attribute. When the state determiner cannot find the corresponding directed connection in the standard connected edge set, it determines the adjacent pair as a broken edge and writes the reason "broken due to no direct connection" into the path attribute.

[0204] If a directed direct connection exists, the state determiner compares the order offset of the two endpoints, the order consistency tolerance threshold, and the severe misorder threshold. If the order offset of the two endpoints does not exceed the order consistency tolerance threshold, it is determined to be a normal connection edge; if it reaches or exceeds the severe misorder threshold, it is determined to be a misordered connection edge (severe); if it is between the two thresholds, it is determined to be a misordered connection edge (mild).

[0205] It should be noted that the judgment result is synchronously written to the path attribute items of adjacent pairs, including the start and end node numbers, connection status, reason for misorder or break, order offset, and whether it contains missing insertion flags, and the student's answer path attributes are output.

[0206] Furthermore, all adjacent pairs identified as fractures are extracted to generate a fracture location list. The fracture location list records the start point, end point, adjacent context position, and fracture cause of each fracture, and outputs the fracture location list.

[0207] Based on the list of broken edge locations and the set of connected edges in the standard cognitive path graph, candidate path sequences are generated by enumerating within the upper bound of the maximum path length. When a candidate path sequence cannot be generated, an event of insufficient candidates is constructed and the maximum path length is widened, and candidate paths are enumerated again.

[0208] Furthermore, the candidate path enumerator extracts the start and end points of each fracture in the fracture location list, expands forward within the set of connecting edges of the standard cognitive path graph according to the constraint of not exceeding the upper bound of the maximum path length, and generates a directed candidate path sequence.

[0209] The generated candidate path sequence, step count, and logical priority labels for each edge are output to the candidate set temporary storage area. If the candidate path enumerator breaks at a certain point and expands according to the upper bound of the maximum path length without generating any candidate paths, the candidate path enumerator immediately constructs a "candidate insufficiency" event message and writes the message to the event queue.

[0210] Furthermore, after receiving the "insufficient candidates" event, the parameter manager only temporarily widens the upper bound of the maximum path length to four steps at the break point, and returns the widened upper bound of the maximum path length to the candidate path enumerator. The candidate path enumerator then performs the enumeration again and outputs the new candidate path sequence to the candidate set temporary storage area.

[0211] The path selector prioritizes the path with the shortest number of steps, combines logical priority and standard order to generate a candidate priority table, fills in the edges or retains the breaks based on the candidate priority table, updates the student's answer path and attributes, and outputs the graph residual scoring structure graph based on the colored node list.

[0212] Furthermore, the path selector sorts the candidate paths at each break point in the candidate set temporary storage area by comparing lengths and then priorities. First, it compares the number of steps of each candidate path and ranks the candidate with the fewest steps first. When there are multiple candidate paths with the same number of synchronizations, it continues to count the number of edges with the highest logical priority in each candidate path and ranks the ones with more edges first. If the logical priorities are still tied, it sorts the candidate paths in a dictionary order according to their appearance in the standard cognitive path graph and outputs the sorting result as a priority list of candidate paths at the break point.

[0213] It should be noted that the path selector reads the candidate path priority list and makes a decision on whether to patch the edge at the break. When there is at least one candidate path, the candidate with the highest ranking is selected as the patching path. The node sequence of the candidate path with the highest ranking is inserted into the corresponding break position in the student's answer path, and the "edge patched, candidate path used, number of steps and logical priority description" is written into the student's answer path attribute. When there is no candidate path, the break remains unchanged and the description "edge not patched, due to insufficient candidates" is written into the student's answer path attribute.

[0214] It should be noted that after each insertion of a patch or preservation of a break, the processing result is immediately written to the log buffer, and the student's answer path after the break is filled and the updated student answer path attributes are output.

[0215] It should be noted that the updated student answer path attributes include the connection status of each pair of adjacent nodes, whether to add edges, and the description of edge addition.

[0216] Furthermore, a graph container for residual scoring structure is created and divided into node area, path area and edge patching area. The student node list is traversed, and a node record is written for each student's knowledge point. The node number, color, order level label, order offset, matching status, scoring level label, whether it is missing or inserted, whether it is redundant, and the order of appearance are written into the node area.

[0217] It should be noted that the order level labels in the residual scoring structure graph are only used for path attributes and break / edge repair determination; the scoring level labels are not modified, and the interpretation and calculation are completed in step five.

[0218] Furthermore, iterate through the student's answer path attributes, write a path record for each pair of adjacent nodes, and write the start and end points, connection status, reason explanation, and whether it was generated by edge supplementation into the path area.

[0219] Furthermore, for the broken sections where the edges have been patched, a patching record is written, and the "patching node sequence and location description" is associated back with the path record.

[0220] It should be noted that the residual scoring structure graph supports functions such as graphical display, structural comparison, and explanatory text generation, and can be directly called by the scoring output module, providing structured residual information support for scoring and report generation.

[0221] S5. Based on the graph residual scoring structure graph, combined with the student knowledge point node scoring annotation information, calculate the structured score, generate the scoring result structure and the explanatory feedback structure, and output the final scoring report.

[0222] Based on the residual scoring structure graph, combined with the student's knowledge point node scoring annotation information, a structured score is calculated using a fixed score rule, and a structured feedback group is generated.

[0223] Furthermore, based on the graph residual scoring structure graph and the set of student knowledge point node scoring annotation information, a structured score is calculated.

[0224] It should be noted that complex rules such as function superposition or floating-point multiplication are not used; instead, a simple fixed score rule is adopted.

[0225] Furthermore, based on the graph residual scoring structure graph, a score is assigned to each knowledge point node in the graph residual scoring structure graph.

[0226] Furthermore, based on the rating level labels determined in the set of rating labels for student knowledge point nodes, a corresponding score is assigned to each student's knowledge point node.

[0227] If the rating level is marked as "full marks", the score for this node is 2 points. If the rating level is marked as "partial marks", the score is 1 point; if the rating level is marked as "no marks", the score is 0 points. The scores for each student's knowledge point node are summarized to calculate the total score.

[0228] To facilitate standardized output, the total score is converted into a structured score with a maximum score of 100.

[0229] Furthermore, by utilizing the color-coded student knowledge point nodes in the residual scoring structure graph, the type of structural problem is determined, and a structured feedback group with clear direction is generated, which focuses on expressing the location and type of problems in the student's answers at the structural level.

[0230] Furthermore, the structural problem types are obtained from the color-coded student knowledge point node annotation results.

[0231] It should be noted that the color-coded student knowledge point node annotation results specifically include node color, node order level, node rating level, missing node list, and redundant node list.

[0232] The judgment rule is based on the results of the students' knowledge point node annotations with color markings.

[0233] Furthermore, if a node's rating is marked as perfect, the structural problem type is "correct expression". If a node's rating is marked as slightly out of order or severely out of order, the structural problem type is "order deviation". If a node is red, the structural problem type is "missing expression". If a node is gray, the structural problem type is "redundant expression".

[0234] Furthermore, structured feedback groups , is represented as:

[0235] ;

[0236] It should be noted that the node content, location-related information, and rating level are derived from the rating level label.

[0237] It should be noted that the generated structured feedback group can be used as input and exported as JSON, tables, or label prompts on graphs, etc., to facilitate human-computer interaction or teacher review, thus completing the entire process from structured annotation fields to reusable feedback formats.

[0238] Based on structured scores and structured feedback groups, a scoring result structure and an explanatory feedback structure are generated.

[0239] Furthermore, a scoring result structure is generated based on the total score, structured score, and score list for each node.

[0240] Based on structured feedback groups, feedback units corresponding to all nodes are collected, thereby generating an interpretive feedback structure, represented as:

[0241] ;

[0242] in, This indicates an explanatory feedback structure. Indicates a structured feedback group. This represents the residual score structure graph. "For all" is a logical quantifier. This indicates that it is necessary to traverse every node in the residual scoring structure graph. This ensures that explanatory feedback for all nodes is generated and incorporated into the explanatory feedback structure.

[0243] Based on the scoring result structure and the explanatory feedback structure, a final scoring report is generated and used for student feedback and teacher review scenarios.

[0244] It should be noted that the scoring result structure and explanatory feedback structure can be applied to student feedback scenarios, teacher review scenarios, and interactive presentation scenarios, thereby supporting a complete data application process after intelligent grading.

[0245] This embodiment also provides a data processing system based on intelligent grading, including:

[0246] The standard graph construction module is used to parse the standard answer text, generate a set of standard knowledge point nodes and a set of connecting edges, assign logical priorities, calculate node weights, and generate a standard cognitive path graph.

[0247] The student knowledge graph generation module is used to parse student answer text, generate candidate nodes for student knowledge points and semantically map them with standard knowledge point nodes to construct a student cognitive path graph.

[0248] The scoring and annotation module is used to compare the standard cognitive path map and the student cognitive path map, analyze the matching status and order offset of student knowledge point nodes, and generate a set of scoring and annotation information for student knowledge point nodes.

[0249] The residual graph construction module is used to verify the standard cognitive path graph, the student cognitive path graph, and the set of student knowledge point scoring annotation information, to annotate the node status and insert missing nodes, to perform path mapping and break detection, to enumerate and select candidate edges at the break points, and to construct the graph residual scoring structure graph.

[0250] The scoring and feedback module, based on the residual scoring structure graph and combined with student knowledge point node scoring annotations, calculates structured scores, generates a scoring result structure and an explanatory feedback structure, and outputs the final scoring report.

[0251] This embodiment also provides a computer device applicable to the data processing method based on intelligent grading, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the data processing method based on intelligent grading as proposed in the above embodiment.

[0252] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0253] In summary, this invention employs the following methods: parsing standard answers using natural language processing, extracting standard knowledge point nodes and connecting edges, assigning logical priorities and calculating node weights, and constructing a standard cognitive path graph; parsing student answers, mapping student knowledge point nodes through semantic matching, and constructing a student cognitive path graph; comparing the two graphs, generating scoring labels by matching node states and order offsets; detecting and completing missing nodes and broken paths, and constructing a residual scoring structure graph; calculating structured scores, and generating a report with explanatory feedback. The system comprises five corresponding modules, and the device includes a memory and a processor. This invention improves the accuracy and interpretability of grading, and the output report is adapted to student feedback and teacher review, making it suitable for automated grading in educational informatization.

[0254] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A data processing method based on intelligent grading, characterized in that: include, A standard cognitive path graph is constructed based on the standard answer text. The text is parsed using natural language processing tools to construct a set of nodes and a set of connecting edges for standard knowledge points, assign logical priorities, calculate the weights of standard knowledge point nodes, and generate the standard cognitive path graph. Using the standard cognitive path map as a reference, the student's answer text is analyzed and semantic matching is performed to map the student's knowledge point nodes to the standard knowledge point nodes, thus constructing the student's cognitive path map. Based on the standard cognitive path map and the student cognitive path map, the matching status and order offset of student knowledge point nodes are compared and analyzed to generate a set of student knowledge point node scoring annotation information. Verify the standard cognitive path graph, the student cognitive path graph, and the set of student knowledge point scoring annotation information; annotate the node status and insert missing nodes; perform path mapping and break detection; enumerate and select candidate edges at break points; and construct the graph residual scoring structure graph. Based on the graph residual scoring structure graph, combined with the student knowledge point node scoring annotation information, the structured score is calculated, the scoring result structure and the explanatory feedback structure are generated, and the final scoring report is output. The execution path mapping and fracture detection involves enumerating and selecting candidate edges at fracture points to construct a graph residual scoring structure graph. The specific steps are as follows: Traverse the student cognitive path graph after inserting the missing node, generate the student answer path sequence, register adjacent pairs, determine the attributes of the student answer path, and generate a list of broken edge locations. Within the upper bound of the maximum path length, candidate path sequences are generated by enumerating. When a candidate path sequence cannot be generated, an event of insufficient candidates is constructed and the maximum path length is widened, and candidate paths are enumerated again. The path selector prioritizes the path with the shortest number of steps, combines logical priority and standard order to generate a candidate priority table, performs edge patching or preserves breaks, updates the student's answer path and attributes, and outputs the graph residual scoring structure graph.

2. The smart grading based data processing method of claim 1, wherein: The specific steps for constructing the node set and connection edge set of the standard knowledge points are as follows: Perform semantic structure parsing to generate a standard set of sentences and a standard dependency syntax tree; Phrases with grammatical rationality and subject knowledge carrying capacity are extracted from the standard dependency syntax tree and encapsulated into standard knowledge point nodes, which are then mapped to a set of standard knowledge point nodes. Extract the original sentence number to which the standard knowledge point node belongs from the standard knowledge point node set, and construct the connection edge set of the standard knowledge point by combining the relative position of the standard knowledge point node in the standard answer text and the semantic logical relationship of the dependency syntax tree.

3. The data processing method based on intelligent grading according to claim 2, wherein: The process of generating a standard cognitive path graph involves determining the connection edge type of standard knowledge points, assigning logical priorities to the connection edges of standard knowledge points, calculating the node weights of standard knowledge points, and constructing a standard cognitive path graph.

4. The data processing method based on intelligent grading according to claim 3, characterized in that: The process involves using a standard cognitive path graph as a reference, parsing student responses and performing semantic matching to map student knowledge point nodes to standard knowledge point nodes, thus constructing a student cognitive path graph. The specific steps are as follows: Perform semantic structure parsing of student response texts, generate a set of student sentences and a student dependency syntax tree, extract phrases to construct and filter preliminary semantic units, and generate an initial set of student candidate knowledge point nodes; Preliminary semantic units are matched with the standard knowledge point vocabulary to generate preliminary semantic mapping relationships and associate them with student candidate knowledge point nodes, thereby generating a set of student candidate knowledge point nodes. Analyze the features of student candidate knowledge points and combine them to generate semantic representations of student candidate knowledge point nodes; Calculate the semantic similarity score, compare the semantic similarity score with the set similarity threshold, and generate a set of mapping results between student candidate knowledge point nodes and standard knowledge point nodes, as well as an updated set of student knowledge point nodes; Construct logical dependencies between student knowledge point nodes, generate a set of connecting edges for student knowledge points, and label the connecting edge types of student knowledge points to build a student cognitive path graph.

5. The data processing method based on intelligent grading according to claim 4, characterized in that: The comparative analysis of student knowledge point node matching status and order offset generates a set of student knowledge point node rating annotation information. The specific steps are as follows: Based on the semantic mapping results of student knowledge point nodes, the correspondence between student knowledge point nodes and standard knowledge point nodes is confirmed, the matching status is marked, and the mapping relationship between student knowledge point nodes and standard knowledge point nodes is generated. The order offset is calculated by extracting the answer order index and the corresponding standard knowledge point node order number from the successfully matched student knowledge point nodes. Map the sequential offset of each successfully matched student knowledge point node to a sequential level label; Based on the sequential level labels, combined with the mapping relationship between student knowledge point nodes and standard knowledge point nodes, and the rating level annotations, a rating triplet for each student knowledge point node is generated. By integrating the rating triplets of all student knowledge point nodes, a set of student knowledge point node rating annotation information is generated. 6.The smart grading based data processing method of claim 5, wherein: The verification standard cognitive path map, student cognitive path map, and student knowledge point scoring annotation information set are used to annotate node status and insert missing nodes. The specific steps are as follows: Perform consistency checks on the standard cognitive path map, the student cognitive path map, and the set of student knowledge point node scoring annotation information, generate independent sessions, create event queues and log buffers, and set an upper bound for the maximum path length. Each student's knowledge point node is labeled with its color and status. For missing nodes, virtual nodes are inserted and missing events are reported. The maximum path length is tightened, and a student cognitive path map with missing nodes and a list of colored nodes are generated.

7. The data processing method based on intelligent grading according to claim 6, characterized in that: The specific steps for calculating structured scores, generating a scoring result structure and an explanatory feedback structure, and outputting a final scoring report are as follows: Based on the graph residual scoring structure graph, combined with the student knowledge point node scoring annotation information, a structured score is calculated using a fixed score rule and a structured feedback group is generated. Based on structured scores and structured feedback groups, a scoring result structure and an explanatory feedback structure are generated; Based on the scoring result structure and the explanatory feedback structure, a final scoring report is generated and used for student feedback and teacher review scenarios.

8. A data processing system based on intelligent grading, based on the data processing method based on intelligent grading according to any one of claims 1 to 7, characterized in that: include, The standard graph construction module is used to parse the standard answer text, generate a set of standard knowledge point nodes and a set of connecting edges, assign logical priorities, calculate node weights, and generate a standard cognitive path graph. The student knowledge graph generation module is used to parse student answer text, generate candidate nodes for student knowledge points, and perform semantic mapping with standard knowledge point nodes to construct a student cognitive path graph. The scoring and annotation module is used to compare the standard cognitive path map and the student cognitive path map, analyze the matching status and order offset of student knowledge point nodes, and generate a set of student knowledge point node scoring and annotation information. The residual graph construction module is used to verify the standard cognitive path graph, the student cognitive path graph, and the student knowledge point scoring annotation information set, annotate the node status and insert missing nodes, perform path mapping and break detection, enumerate and select candidate edges at the break points, and construct the graph residual scoring structure graph. The scoring and feedback module, based on the residual scoring structure graph, combined with the student's knowledge point node scoring annotation information, calculates the structured score, generates the scoring result structure and explanatory feedback structure, and outputs the final scoring report.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the data processing method based on intelligent grading as described in any one of claims 1 to 7.

Images