An english grammar learning path generation method based on knowledge graph reasoning

Abstract

This invention discloses a method for generating English grammar learning paths based on knowledge graph reasoning, relating to the field of smart education technology. The method includes acquiring relevant data on English grammar learning and preprocessing it to obtain basic grammar learning data; based on the rule deviation direction in the three-label rule deviation state vector, performing reverse reasoning of error cognition along the misuse relation edges in the English grammar knowledge graph to form candidate error cognition types, and combining this with historical answer data to evaluate the confidence level of error cognition and determine the target error cognition type; based on the target error cognition type, performing correction path reasoning along the correction relation edges in the English grammar knowledge graph to obtain candidate correction paths, evaluating the path cost, determining the target correction path, and generating the English grammar learning path. This invention, through the target error cognition type, can summarize and reflect the deviation relationship between examination rules, answer rules, and response rules, improving the clarity of learners' weak points.

Description

Technical Field

[0001] This invention relates to the field of smart education technology, and in particular to a method for generating English grammar learning paths based on knowledge graph reasoning. Background Technology

[0002] With the development of intelligent education, learning analytics, and natural language processing technologies, English grammar learning is gradually evolving from standardized knowledge instruction to personalized diagnosis and adaptive learning. Online learning platforms can collect answer records, incorrect question data, and learning behavior information, and combine them with grammar rule recognition, semantic analysis, and knowledge organization technologies to provide a data foundation for grammar ability assessment and learning resource recommendation.

[0003] Current methods for generating English grammar learning paths often rely on test-taking accuracy, incorrect knowledge points, or pre-set learning units for recommendations. These methods primarily focus on reinforcing knowledge points at the result level, failing to differentiate the specific cognitive reasons behind grammatical errors. When there are complex discrepancies between the test focus, standard answers, and actual responses, errors from different causes are easily grouped together, resulting in insufficiently targeted learning paths. Summary of the Invention

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

[0005] Therefore, this invention provides a method for generating English grammar learning paths based on knowledge graph reasoning to solve the problem that it is difficult to accurately represent the causes of errors in English grammar learning, resulting in insufficient matching of learning content recommendations.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: This invention provides a method for generating English grammar learning paths based on knowledge graph reasoning, which includes acquiring relevant data on English grammar learning and preprocessing it to obtain basic data for grammar learning; Based on the basic data of grammar learning, we identify grammar rule nodes, error cognition nodes, and correction learning nodes, construct misuse relationship edges between grammar rule nodes and error cognition nodes, and correction relationship edges between error cognition nodes and correction learning nodes, thus forming an English grammar knowledge graph. Acquire grammar question data, current answer data, and historical answer data; identify examination rule tags and answer rule tags from the grammar question data; identify answer rule tags from the current answer data; and perform contextual analysis on the grammar question data to obtain question context information. We perform rule deviation analysis on the examination rule labels, answer rule labels, and answer rule labels, and combine this with the contextual information of the question to generate a three-label rule deviation state vector that includes the direction of rule deviation. Based on the rule deviation direction in the three-label rule deviation state vector, error cognition retrieval and reasoning are performed along the misuse relation edge in the English grammar knowledge graph to form candidate error cognition types. Then, error cognition confidence is evaluated in combination with historical answer data to determine the target error cognition type. Based on the target error cognitive type, the correction path reasoning is performed along the correction relation edges in the English grammar knowledge graph to obtain candidate correction paths. The path cost is then evaluated to determine the target correction path and generate an English grammar learning path.

[0007] As a preferred embodiment of the English grammar learning path generation method based on knowledge graph reasoning described in this invention, the specific steps for obtaining the basic grammar learning data are as follows: Acquire English grammar rule texts, grammar question data, question analysis, learning records, and incorrect question records, and perform data cleaning and format standardization to obtain standardized grammar learning data; By analyzing grammatical elements and organizing correction-related information from standardized grammar learning data, basic grammar learning data is obtained.

[0008] As a preferred embodiment of the English grammar learning path generation method based on knowledge graph reasoning described in this invention, the specific steps for forming an English grammar knowledge graph are as follows: Extract grammar rule labels, misuse information, erroneous cognition information, correction rule information, and learning content information from basic grammar learning data, and perform synonym merging and category normalization to form grammar rule nodes, erroneous cognition nodes, and correction learning nodes. Based on the correspondence between grammar rule labels, misuse performance information, and incorrect cognitive information, misuse triples are formed, and misuse relationship edges are constructed between grammar rule nodes and incorrect cognitive nodes; Based on the correspondence between erroneous cognitive information, correction rule information, and learning content information, a correction triplet is formed, and a correction relationship edge is constructed between erroneous cognitive nodes and correction learning nodes. By associating and storing grammar rule nodes, error cognition nodes, correction learning nodes, misuse relation edges, and correction relation edges, an English grammar knowledge graph is formed.

[0009] As a preferred embodiment of the English grammar learning path generation method based on knowledge graph reasoning described in this invention, the steps of obtaining the examination rule label, the answer rule label, and the answer rule label respectively refer to acquiring grammar question data and current answer data, extracting the question stem, question analysis, and standard answer from the grammar question data, and extracting the user answer from the current answer data; identifying the examination rule label based on the question stem and question analysis, and placing the user answer and standard answer into the corresponding positions in the question stem to perform grammar rule identification, thereby obtaining the answer rule label and the answer rule label.

[0010] As a preferred embodiment of the English grammar learning path generation method based on knowledge graph reasoning described in this invention, the method of obtaining the question context information refers to performing context analysis on the question stem, question analysis and standard answer, extracting syntactic structure information, grammatical trigger information and the grammatical position of the standard answer, and associating the syntactic structure information, grammatical trigger information and the grammatical position of the standard answer to form the question context information.

[0011] As a preferred embodiment of the English grammar learning path generation method based on knowledge graph reasoning described in this invention, the specific steps for generating a three-label rule deviation state vector containing rule deviation directions are as follows: Using the examination rule tags as rule references, the answer rule tags and the answer rule tags are compared with the examination rule tags in terms of rule category and grammatical function to form a three-tag rule reference relationship; Based on the three-label rule reference relationship, the deviation status of the answer rule label relative to the examination rule label and the matching status of the answer rule label relative to the examination rule label are verified to form the three-label rule deviation relationship; Based on the three-label rule deviation relationship, the difference between the answer rule label and the answer rule label is mapped to the syntactic structure information and grammatical trigger information in the question context information to form the rule deviation direction; Based on the direction of the rule deviation, the relationship between the three-label rule deviations and the contextual information of the question are combined and encoded to generate a three-label rule deviation state vector containing the direction of the rule deviations.

[0012] As a preferred embodiment of the English grammar learning path generation method based on knowledge graph reasoning described in this invention, the specific steps for forming candidate error cognitive types are as follows: Read the rule deviation direction from the three-label rule deviation state vector, and match the rule deviation direction with the examination rule label and the answer rule label to form the misuse link retrieval conditions; Based on the retrieval criteria for misuse links, misuse links associated with the examination rule tags and answer rule tags are retrieved along the misuse relationship edges in the English grammar knowledge graph to form candidate misuse links; Based on the answer rule tags, the rule pointers corresponding to the candidate misuse links are checked to determine the rule consistency status between the candidate misuse links and the examination rule tags. Based on the question context information, the matching status between the candidate misuse links and the question context information is checked to form a valid misuse link. Based on the erroneous cognitive nodes associated with the effective misuse links, the erroneous cognitive information corresponding to the erroneous cognitive nodes is classified to form candidate erroneous cognitive types.

[0013] As a preferred embodiment of the English grammar learning path generation method based on knowledge graph reasoning described in this invention, the specific steps for determining the target error cognitive type are as follows: Based on the direction of rule deviation, the examination rule label, and the answer rule label, filter historical records of deviations in the same direction that are consistent with the current answer deviation from the historical answer data; Based on historical records of similar deviations, the misuse relation edges corresponding to candidate error cognitive types are matched with historical records of similar deviations. The number of occurrences of similar deviations, the number of consecutive occurrences, and the number of recurrences after corrective learning are counted to form frequency support values, consecutive support values, and recurrence support values. Based on frequency support values, continuity support values, and recurrence support values, the confidence level of candidate error cognition types is evaluated to obtain the error cognition confidence level corresponding to each candidate error cognition type; Candidate error cognitive types are sorted according to their error cognitive confidence. When the error cognitive confidence is the same, they are sorted in the following order: recurrence support value, continuous support value, frequency support value, and the order in which the candidate error cognitive types were generated. The candidate error cognitive type ranked first is determined as the target error cognitive type.

[0014] As a preferred embodiment of the English grammar learning path generation method based on knowledge graph reasoning described in this invention, the specific steps for obtaining candidate correction paths are as follows: Based on the error cognitive information in the target error cognitive type, the matching error cognitive node is determined in the English grammar knowledge graph and used as the starting point for the correction path retrieval; Starting from the starting point of the correction path retrieval, the associated correction learning nodes are retrieved along the correction relationship edges. The retrieved path is then verified by combining the examination rule label, answer rule label, and rule deviation direction to obtain candidate correction paths.

[0015] As a preferred embodiment of the English grammar learning path generation method based on knowledge graph reasoning described in this invention, the specific steps for generating the English grammar learning path are as follows: The candidate correction paths are matched and verified with the target error cognition type, rule deviation direction, examination rule label and answer rule label to form candidate correction path verification information; Based on the verification information of the candidate correction path, the path cost of the candidate correction path is evaluated according to the correction learning nodes and historical answer data in the candidate correction path, and the cost value of the candidate correction path is formed. Candidate correction paths are sorted according to their cost value, and the target correction path is determined from the candidate correction paths based on the direction of rule deviation, thus generating an English grammar learning path.

[0016] The beneficial effects of this invention are as follows: By determining the target error cognitive type based on the rule deviation direction in the three-label rule deviation state vector, learners' answer deviations can be transformed into cognitively oriented expression of error causes, establishing a stable connection between rule deviation representation and error cause localization. The target error cognitive type can summarize and reflect the deviation relationship between examination rules, answer rules, and answer rules, improving the clarity of learners' weak points and enhancing the consistency and reliability of the learning content matching criteria. Attached Figure Description

[0017] 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.

[0018] Figure 1 This is a flowchart of a method for generating English grammar learning paths based on knowledge graph reasoning.

[0019] Figure 2 A flowchart for generating a three-label rule deviation state vector.

[0020] Figure 3 A flowchart for determining the type of target misperception.

[0021] Figure 4 This is a flowchart of the English grammar learning path. Detailed Implementation

[0022] 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.

[0023] 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.

[0024] 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 present invention. The phrase "in one embodiment" appearing in different places in 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.

[0025] Reference Figures 1-4This is one embodiment of the present invention, which provides a method for generating English grammar learning paths based on knowledge graph reasoning, including the following steps: S1: Obtain relevant data on English grammar learning and preprocess it to obtain basic data for grammar learning.

[0026] S1.1: Obtain English grammar rule text, grammar question data, question analysis, learning records and wrong question records, and perform data cleaning and format standardization to obtain standardized grammar learning data.

[0027] Acquire English grammar rule texts, grammar question data, question analysis, learning records, and incorrect question records, retaining the original content, original sorting order, and acquisition time of the acquired data.

[0028] Data involving user identifiers in learning records and incorrect answer records undergoes authorization verification and anonymization. Data without authorization or unable to complete anonymization will not be processed further. All learning records, incorrect answer records, and user identifiers are obtained with the user's consent and used for legitimate purposes.

[0029] Add a data source number and a data sequence number to the acquired data according to the data acquisition batch and the original sorting order, and combine the data source number and the data sequence number into a source data identifier.

[0030] Extract grammar rule text identifiers and grammar rule descriptions from English grammar rule texts; extract question identifiers, question stems, and answer content from grammar question data; extract question identifiers, question stem related content, and answer basis content from question analysis; extract learning record user identifiers, learning object type, learning object identifiers, learning time, answer content, answer result, and correction learning records from learning records; extract user identifiers, question identifiers, incorrect answer content, standard answer content, and error occurrence time from incorrect question records, and unify the de-identified learning record user identifiers and user identifiers into a single de-identified user identifier.

[0031] Based on the question identifier, learning object identifier, grammar rule text identifier, and source data identifier, establish the correspondence between grammar question data, question analysis, learning records, error records, and English grammar rule text.

[0032] When multiple candidate correspondences exist, the correspondences are determined in the following order: data source number is the same, data sequence number difference is the smallest, and acquisition time difference is the smallest. When multiple candidate correspondences still exist, the corresponding data is marked as data to be reviewed, and the data to be reviewed will not participate in the subsequent parsing of grammatical elements.

[0033] Based on the correspondence, the missing answer content of the grammar question data is completed from the answer content in the question analysis, the missing incorrect answer content in the wrong question record is completed from the answer content in the learning record, and the missing standard answer content in the wrong question record is completed from the answer content in the grammar question data; when data that cannot be uniquely completed is marked as data to be reviewed, the data to be reviewed will not participate in the subsequent grammar element analysis.

[0034] Remove leading and trailing whitespace, duplicate spaces, tabs, web page tags, bullets, invisible control characters, and abnormal line breaks from each completed data entry. Convert full-width English letters, numbers, and punctuation marks to half-width characters, while preserving individual spaces between words and their original capitalization in the English text.

[0035] The data is standardized by removing duplicates from the following information: grammar rules description, question stem, answer, answer basis, anonymized user identifier, learning target identifier, learning time, answer content, answer result, incorrect answer content, and error occurrence time.

[0036] S1.2: Perform grammatical element analysis and correction association information organization on the standardized grammar learning data to obtain basic grammar learning data.

[0037] Based on the grammar rule descriptions, question analyses, and answer bases in the standardized grammar learning data, we extract grammar terminology names, definitions, English expressions, common aliases, and grammar knowledge point names. Then, assuming consistency in grammar knowledge point names, we merge records with consistent grammar terminology names, definitions, English expressions, or common aliases to obtain an English grammar terminology table.

[0038] The English grammar glossary includes the names of grammar terms, their definitions, English expressions, common alternative names, and names of grammar points.

[0039] Based on the English grammar terminology table, the grammar rule descriptions, question stems, answer content, and answer basis content in the standardized grammar learning data are organized into rules. Word form change rules, part-of-speech collocation rules, syntactic relationship rules, grammatical structure content, grammatical application content, and corresponding grammar rule tags are extracted to form word form rule tables, part-of-speech collocation rule tables, syntactic relationship rule tables, and rule tag matching rule tables.

[0040] By comparing the incorrect answers and standard answers in the error record, the types and features of the differences are extracted. The types and features of the differences are then correlated with information on misuse and incorrect cognition, forming a misuse performance matching rule table.

[0041] Using a glossary of English grammar terms, a table of word forms, a table of collocation rules, a table of syntactic relations, and a table of rule tag matching rules, we perform word segmentation, part-of-speech tagging, word form restoration, and syntactic relation matching on the grammar rule descriptions, question stems, answer content, and answer basis content to determine the corresponding grammatical structure content, applicable grammar content, and grammar rule tags, thus obtaining grammar parsing data.

[0042] The grammar parsing data includes grammar structure content, applicable grammar content, grammar rule tags, question identifiers, and source data identifiers.

[0043] Based on the learning object type and learning object identifier, the learning records are associated with English grammar rule texts and grammar question data respectively, and the correction learning records associated with the corresponding grammar rule tags are collected according to the grammar rule tags.

[0044] Based on the question identifier, the corresponding grammar question data and question analysis are read. The incorrect answers in the wrong question record are compared with the standard answer in terms of word form changes, parts of speech collocation, syntactic relations and grammatical application conditions to determine the difference type, difference characteristics, position of the incorrect answer and position of the standard answer, and thus obtain the answer difference data.

[0045] The difference types and characteristics in the answer difference data are matched with the difference types and characteristics in the misuse performance matching rule table. When the match is successful, the misuse performance information and erroneous cognition information in the corresponding misuse performance matching rule table record are identified as the misuse performance information and erroneous cognition information corresponding to the wrong question record. When the match is unsuccessful, the corresponding wrong question record is marked as data to be reviewed and will not participate in subsequent processing before review.

[0046] The answer in the question analysis is identified as the correction rule information. Based on the grammar rule tag corresponding to the correction rule information, the grammar rule description content under the same grammar rule tag is matched in the English grammar rule text. The matched grammar rule description content is identified as the learning content information.

[0047] When multiple candidate learning content information exist, the unique learning content information is determined according to the following rules: consistent grammatical structure, consistent grammatical application, smallest data sequence number, and earliest acquisition time.

[0048] Historical answer data is generated in chronological order based on the anonymized user identifier, question identifier, error occurrence time, answer difference data, misuse behavior information, error cognition information, corresponding grammar rule tags, answer results, and correction learning records.

[0049] Establish a correspondence between the English grammar rule text, grammar question data, question analysis, grammar analysis data, answer difference data, misuse information, erroneous cognition information, correction rule information, learning content information, and historical answer data under the same grammar rule label to form basic data for grammar learning.

[0050] S2: Based on the basic data of grammar learning, determine the grammar rule nodes, error cognition nodes, and correction learning nodes, construct the misuse relationship edges between grammar rule nodes and error cognition nodes, and the correction relationship edges between error cognition nodes and correction learning nodes, to form an English grammar knowledge graph.

[0051] S2.1: Extract grammar rule labels, misuse information, erroneous cognition information, correction rule information, and learning content information from the basic grammar learning data, and perform synonym merging and category normalization to form grammar rule nodes, erroneous cognition nodes, and correction learning nodes.

[0052] Using the grammar rule tags in the basic grammar learning data as candidate node names, the corresponding grammar structure content, applicable grammar content, and grammar knowledge point names are associated with the same candidate record to form grammar rule node candidate data.

[0053] Using the erroneous cognitive information in the basic grammar learning data as candidate node names, the corresponding misuse performance information, answer difference data and corresponding grammar rule labels are associated with the same candidate record to form erroneous cognitive node candidate data.

[0054] Using the learning content information in the basic grammar learning data as candidate node names, the corresponding correction rule information and grammar rule labels are associated with the same candidate record to form candidate data for correction learning nodes.

[0055] Based on the English grammar terminology, error cognition information, and learning content information, the candidate node names in the candidate data of grammar rule nodes, error cognition nodes, and error correction learning nodes are merged into synonyms.

[0056] When a candidate node name matches a grammatical term name, English expression, or common alias in the English grammar terminology table, the corresponding grammatical term name is determined as the merged node name; when a candidate node name does not match the English grammar terminology table, the candidate node name is determined as the merged node name.

[0057] Based on the candidate node names and node attributes in the candidate records, the candidate data of grammar rule nodes, error cognition nodes, and correction learning nodes are subjected to category normalization. Candidate records with the same meaning and the same node category are merged to form grammar rule nodes, error cognition nodes, and correction learning nodes.

[0058] Node identifiers are generated based on the merged node names, node categories, and grammar knowledge point names. The node identifiers, node names, node categories, grammar knowledge point names, and node attributes are then linked and recorded to form grammar rule nodes, error perception nodes, and correction learning nodes. The node attributes of grammar rule nodes include grammar structure content and grammar application content. The node attributes of error perception nodes include misuse performance information and answer difference data. The node attributes of correction learning nodes include correction rule information and learning content information.

[0059] S2.2: Based on the correspondence between grammar rule labels, misuse performance information, and error cognition information, form misuse triples and construct misuse relationship edges between grammar rule nodes and error cognition nodes.

[0060] Based on the misuse performance information and erroneous cognitive information corresponding to the same grammar rule label in the basic grammar learning data, the grammar rule node corresponding to the grammar rule label is determined as the starting object of the misuse triple, the misuse performance information is determined as the relation content of the misuse triple, and the erroneous cognitive node corresponding to the erroneous cognitive information is determined as the ending object of the misuse triple, thus generating the misuse triple.

[0061] Match the grammar rule labels in the misuse triples with the node names of the grammar rule nodes, and match the erroneous cognitive information in the misuse triples with the node names of the erroneous cognitive nodes.

[0062] When a match is successful, the syntax rule node is determined as the starting node of the misuse relation edge, the erroneous cognition node is determined as the ending node of the misuse relation edge, the misuse performance information is written into the relation attribute of the misuse relation edge, and the misuse relation edge between the syntax rule node and the erroneous cognition node is constructed.

[0063] When a grammar rule label or erroneous cognitive information does not match the corresponding node, the corresponding misuse triplet is marked as data to be reviewed, and misuse relation edges are not constructed before review.

[0064] When multiple misuse relation edges correspond to the same syntax rule node, the same erroneous cognition node, and the same misuse performance information, the multiple misuse relation edges are merged into one misuse relation edge.

[0065] S2.3: Based on the correspondence between erroneous cognitive information, correction rule information, and learning content information, form correction triples and construct correction relationship edges between erroneous cognitive nodes and correction learning nodes.

[0066] Based on the error correction rule information and learning content information corresponding to the same error cognitive information in the basic data of grammar learning, the error cognitive node corresponding to the error cognitive information is determined as the starting object of the error correction triple, the error correction rule information is determined as the relation content of the error correction triple, and the error correction learning node corresponding to the learning content information is determined as the ending object of the error correction triple, thus generating the error correction triple.

[0067] Match the erroneous cognitive information in the correction triple with the node name of the erroneous cognitive node, and match the learning content information in the correction triple with the node name of the correction learning node. When a match is successful, the erroneous cognitive node is determined as the starting node of the correction relationship edge, the correction learning node is determined as the ending node of the correction relationship edge, the correction rule information is written into the relationship attribute of the correction relationship edge, and the correction relationship edge between the erroneous cognitive node and the correction learning node is constructed.

[0068] When incorrect cognitive information or learning content information does not match the corresponding node, the corresponding correction triplet is marked as data to be reviewed, and no correction relationship edge is constructed before review.

[0069] When multiple corrective relation edges correspond to the same erroneous cognition node, the same corrective learning node, and the same corrective rule information, the multiple corrective relation edges are merged into one corrective relation edge.

[0070] S2.4: Link and store grammar rule nodes, error cognition nodes, correction learning nodes, misuse relation edges, and correction relation edges to form an English grammar knowledge graph.

[0071] Based on the start and end nodes in the misuse relation edge, the misuse relation edge is associated with the corresponding syntax rule node and error cognition node, and the start node identifier, end node identifier, misuse performance information and relation type are recorded in the misuse relation edge.

[0072] Based on the start and end nodes in the correction relationship edge, the correction relationship edge is associated with the corresponding erroneous cognition node and correction learning node, and the start node identifier, end node identifier, correction rule information, learning content information, and relationship type are recorded in the correction relationship edge. The misused relation edges are stored according to the start node identifier, end node identifier, relation type, and misuse behavior information. The corrective relation edges are stored according to the start node identifier, end node identifier, relation type, and correction rule information. A correspondence between node identifiers and start node identifiers and end node identifiers is established to form an English grammar knowledge graph.

[0073] S3: Obtain grammar question data, current answer data, and historical answer data. Combine the grammar question data and current answer data to identify examination rule tags, answer rule tags, and answer rule tags. Perform contextual analysis on the grammar question data to obtain question context information.

[0074] S3.1 Obtain grammar question data and current answer data, extract question stem, question analysis and standard answer from grammar question data, and extract user answer from current answer data; identify examination rule tags based on question stem and question analysis, and place user answer and standard answer into the corresponding positions in question stem to perform grammar rule recognition, and obtain answer rule tags and answer rule tags.

[0075] The system retrieves the grammar question data associated with the user's current answer, the current answer data, and the historical answer data. It reads the question identifier, question stem, question analysis, and standard answer from the grammar question data, and reads the user's answer, anonymized user identifier, and answer time from the current answer data. It then associates the historical answer records with the question identifier and anonymized user identifier to form the current answer associated data.

[0076] Based on the blank positions in the question stem, the option filling positions in the grammar question data, or the answer reference positions in the question analysis, determine the grammatical position of the standard answer in the question stem; when there are multiple candidate grammatical positions and it cannot be uniquely determined, mark the current answer-related data as data to be reviewed, and do not participate in the generation of the three-label rule deviation state vector before review.

[0077] The standard answer and the user's answer are written in the same grammatical location to form the standard answer context text and the user's answer context text.

[0078] Based on the English grammar terminology table, word form rule table, part-of-speech collocation rule table, syntactic relation rule table, and rule tag matching rule table, word segmentation, part-of-speech tagging, word form restoration, and syntactic relation matching are performed on the question stem, question analysis, standard answer context text, and user answer context text to obtain the current answer grammar analysis data.

[0079] Based on the current grammar analysis data of the answers, the grammar rule tags that match the grammar position and answer basis content in the question stem and question analysis are determined as examination rule tags, the grammar rule tags corresponding to the context text of the user's answer are determined as answer rule tags, and the grammar rule tags corresponding to the context text of the standard answer are determined as answer rule tags. When any tag cannot be uniquely determined, the current answer-related data is marked as data to be reviewed, and it will not participate in the generation of the three-tag rule deviation state vector before review.

[0080] S3.2: Perform contextual analysis on the question stem, question analysis, and standard answer, extract syntactic structure information, grammatical trigger information, and the grammatical position of the standard answer, and associate the syntactic structure information, grammatical trigger information, and the grammatical position of the standard answer to form the question context information.

[0081] Based on the grammatical location of the standard answer, determine the contextual scope that includes the grammatical location of the standard answer in the question stem, and associate the contextual scope with the answer basis content in the question analysis to form a contextual analysis object.

[0082] Based on the syntactic relation rule table, the syntactic components of the context parsing object are divided to determine the lexical positions of the subject, predicate, object, complement, attributive, adverbial and complement, thus obtaining syntactic structure information. When the context parsing object cannot be completely divided into syntactic components, the positions of the identified syntactic components are retained and the missing syntactic components are marked as null values.

[0083] Based on the rule table of rule tags and the answer basis content in the question analysis, the grammatical trigger information is obtained by identifying word form changes, parts of speech collocations, syntactic relations, grammatical structure content and grammatical application content related to the grammatical position of the standard answer in the context analysis object.

[0084] By associating syntactic structure information, grammatical trigger information, the grammatical position of the standard answer, the examination rule tags, and the answer rule tags, the contextual information of the question is formed.

[0085] S4: Perform rule deviation analysis on the examination rule label, the answer rule label, and the answer rule label, and generate a three-label rule deviation state vector containing the direction of rule deviation by combining the context information of the question.

[0086] S4.1: Using the examination rule label as the rule reference, compare the answer rule label and the answer rule label with the examination rule label in terms of rule category and grammatical function to form a three-label rule reference relationship.

[0087] The examination rule tags, answer rule tags, and answer rule tags are read from the current answer grammar analysis data. Combined with the grammar position of the standard answer, the answer basis content in the question analysis, and the grammar trigger information, the grammar knowledge point name, grammar structure content, and grammar application content corresponding to each rule tag are read from the grammar learning basic data.

[0088] When the name of the grammatical knowledge point, the content of the grammatical structure, or the content of the grammatical application corresponding to any rule label cannot be uniquely determined, the current answer-related data is marked as data to be reviewed, and will not participate in the generation of the three-label rule deviation state vector before review.

[0089] The names of grammatical knowledge points and grammatical structures corresponding to the examination rule tags are determined as the reference content for rule categories, and the applicable grammatical content corresponding to the examination rule tags is determined as the reference content for grammatical functions.

[0090] Based on the English grammar terminology table and the rule tag matching rule table, the answer rule tags and the answer rule tags are compared with the rule category reference content for rule category consistency, and then compared with the grammar function reference content for grammar function consistency, to obtain the answer rule category reference result, the answer grammar function reference result, the answer rule category reference result, and the answer grammar function reference result.

[0091] The examination rule label, the answer rule label, the answer rule label, the answer rule category reference result, the answer grammar function reference result, the answer rule category reference result, and the answer grammar function reference result are linked to form a three-label rule reference relationship.

[0092] S4.2: Based on the three-label rule reference relationship, verify the deviation status of the answer rule label relative to the examination rule label, and the matching status of the answer rule label relative to the examination rule label, to form a three-label rule deviation relationship.

[0093] Based on the three-label rule reference relationship, read the answer rule category reference result and the answer grammar function reference result; when both are consistent, the answer rule label is determined to be in an undeviated state relative to the examination rule label; when the answer rule category reference result is inconsistent but the answer grammar function reference result is consistent, it is determined to be in a rule category deviation state; when the answer rule category reference result is consistent but the answer grammar function reference result is inconsistent, it is determined to be in a grammar function deviation state; when both are inconsistent, it is determined to be in a rule category and grammar function deviation state.

[0094] Based on the three-label rule reference relationship, read the answer rule category reference result and the answer grammar function reference result; when both are consistent, determine the answer rule label as a match relative to the examination rule label; when either result is inconsistent, determine it as a mismatch.

[0095] When the answer rule label is in an unbiased state relative to the examination rule label, and the answer rule label is in a matching state relative to the examination rule label, the current answer-related data is marked as data that does not need correction. Data that does not need correction does not participate in the generation of rule deviation direction and erroneous cognition retrieval reasoning.

[0096] The deviation status of the answer rule label from the examination rule label, the matching status of the answer rule label from the examination rule label, the examination rule label, the answer rule label, and the answer rule label are associated to form a three-label rule deviation relationship.

[0097] S4.3: Based on the three-label rule deviation relationship, the difference between the answer rule label and the answer rule label is mapped to the syntactic structure information and grammatical trigger information in the question context information to form the rule deviation direction.

[0098] Based on the deviation relationship of the three-label rule, the deviation status of the answer rule label relative to the examination rule label and the matching status of the answer rule label relative to the examination rule label are read; when the answer rule label is mismatched with the examination rule label, the current answer-related data is marked as data to be reviewed, and it will not participate in the erroneous cognitive retrieval reasoning before review.

[0099] When the answer rule label matches the examination rule label, the grammar knowledge point name, grammar structure content, and applicable grammar content corresponding to the answer rule label are compared with the grammar knowledge point name, grammar structure content, and applicable grammar content corresponding to the answer rule label to determine the difference status; when all three items are consistent, the current answer-related data is marked as data without rule deviation and is not included in the error cognitive retrieval reasoning.

[0100] When at least one of the grammatical knowledge point names or grammatical structure contents is inconsistent, but the applicable content of the grammar is consistent, the difference state is determined to be a rule category difference state; when both the grammatical knowledge point names and grammatical structure contents are consistent, but the applicable content of the grammar is inconsistent, the difference state is determined to be a grammatical function difference state; when at least one of the grammatical knowledge point names or grammatical structure contents is inconsistent, and the applicable content of the grammar is inconsistent, the difference state is determined to be a rule category and grammatical function difference state.

[0101] The standard answer's grammatical location, syntactic structure, and grammatical trigger information are retrieved from the question's context. The difference status is then mapped to the standard answer's grammatical location. When the standard answer's grammatical location belongs to a syntactic component within the syntactic structure information, and the grammatical structure and applicable content corresponding to the difference status are consistent with the grammatical structure and applicable content in the grammatical trigger information, the difference status is determined as a valid difference status. When the difference status cannot be determined as a valid difference status, the current answer-related data is marked as data awaiting review and will not participate in the generation of the three-label rule deviation status vector before review.

[0102] Based on the valid difference status, the grammatical position of the standard answer, syntactic structure information, and grammatical trigger information, a rule deviation direction is formed. The rule deviation direction includes the difference status, the grammatical position of the standard answer, the corresponding syntactic components, the grammatical trigger information, and the deviation content of the answer rule label relative to the answer rule label. S4.4: Based on the direction of the rule deviation, the three-label rule deviation relationship and the question context information are combined and encoded to generate a three-label rule deviation state vector containing the direction of the rule deviation.

[0103] The deviation status of the answer rule label from the examination rule label is encoded. The non-deviation status corresponds to the code value 0, the rule category deviation status corresponds to the code value 1, the grammatical function deviation status corresponds to the code value 2, and the rule category and grammatical function deviation status corresponds to the code value 3, thus obtaining the answer deviation status code value.

[0104] The matching status of the answer rule label relative to the examination rule label is encoded, with a matching status corresponding to the encoding value 1 and a non-matching status corresponding to the encoding value 0, thus obtaining the answer matching status encoding value.

[0105] The differences between the answer rule label and the answer rule label are encoded. The rule category difference corresponds to the encoding value 1, the grammatical function difference corresponds to the encoding value 2, and the rule category and grammatical function difference corresponds to the encoding value 3, thus obtaining the rule difference status encoding value.

[0106] The encoding is performed based on the syntactic structure information, grammatical trigger information, and standard answer grammatical position in the deviation direction of the rules; the corresponding syntactic components in the deviation direction of the rules are encoded according to the fixed order of subject, predicate, object, complement, attributive, adverbial, and complement; the grammatical trigger information is encoded according to the fixed order of word form change, part of speech collocation, syntactic relationship, grammatical structure content, and grammatical application content; the standard answer grammatical position is encoded according to the word position sequence number in the question stem.

[0107] The answer deviation status encoding value, answer matching status encoding value, rule difference status encoding value, syntactic structure information encoding value, grammar trigger information encoding value, and grammar position encoding value are arranged in a fixed field order to generate a three-label rule deviation status vector containing the rule deviation direction.

[0108] The encoded value is used to distinguish different state categories, but does not represent the numerical relationship between different state categories.

[0109] S5: Based on the rule deviation direction in the three-label rule deviation state vector, perform error cognition retrieval and reasoning along the misuse relation edge in the English grammar knowledge graph to form candidate error cognition types, and combine historical answer data to evaluate the confidence of error cognition and determine the target error cognition type.

[0110] S5.1: Read the rule deviation direction from the three-label rule deviation state vector, and match the rule deviation direction with the examination rule label and the answer rule label to form the misuse link retrieval conditions.

[0111] Read the response deviation state encoding value, rule difference state encoding value, syntactic structure information encoding value, grammar trigger information encoding value, and grammar position encoding value from the three-label rule deviation state vector, and read the rule deviation direction associated with the three-label rule deviation state vector.

[0112] By mapping the direction of the rule deviation to the name of the grammatical knowledge point, the content of the grammatical structure, and the applicable content of the grammar corresponding to the test rule label, the rule reference range corresponding to the current answer deviation is determined.

[0113] Based on the direction of rule deviation, the difference status between the answer rule label and the answer rule label, determine the answer deviation content that includes the difference type, difference characteristics and misuse manifestation content.

[0114] The rules reference range, the content of the answer deviation, the encoding value of the grammatical position, the encoding value of the syntactic structure information, and the encoding value of the grammatical trigger information are associated to form the misuse link retrieval conditions.

[0115] S5.2: Based on the misuse link retrieval conditions, retrieve misuse links associated with the examination rule label and the answer rule label along the misuse relationship edge in the English grammar knowledge graph to form candidate misuse links.

[0116] Based on the rule reference range in the misuse link retrieval conditions, the grammar rule nodes corresponding to the examination rule labels are determined in the English grammar knowledge graph, and the grammar rule nodes corresponding to the examination rule labels are determined as the starting points for misuse relation edge retrieval.

[0117] Based on the deviation content of the answer in the misuse link retrieval conditions, retrieve misuse relation edges in the misuse relation edges whose starting node is a syntax rule node and whose misuse performance information in relation attributes is consistent with the deviation content of the answer.

[0118] Read the erroneous cognition node corresponding to the termination node of the misuse relationship edge, and extract the answer difference data associated with the misuse relationship edge from the node attributes of the erroneous cognition node.

[0119] The error answer locations, error types, and error features in the answer difference data are compared with the standard answer grammatical location corresponding to the grammatical location encoding value, the syntactic component corresponding to the syntactic structure information encoding value, and the grammatical trigger content corresponding to the grammatical trigger information encoding value, respectively.

[0120] When the location, type, and characteristics of the incorrect answer all form a consistent correspondence, the corresponding misuse relation edge is determined as a misuse relation edge that has passed the correspondence verification.

[0121] By linking the corresponding misuse relationship edges with grammar rule nodes, misuse performance information, answer difference data, and erroneous cognition nodes, a candidate misuse link will be formed.

[0122] When no misuse relationship edge matching the misuse link retrieval conditions is found, the currently answered related data is marked as data to be reviewed, and will not be involved in the determination of the target error cognition type before review.

[0123] S5.3: Verify the rule pointers corresponding to the candidate misuse links based on the answer rule tags, determine the rule consistency status between the candidate misuse links and the examination rule tags, and verify the matching status between the candidate misuse links and the question context information based on the question context information to form a valid misuse link.

[0124] Based on the grammar knowledge point name, grammar structure content, and applicable grammar content corresponding to the answer rule tags, rule pointing verification is performed on the grammar knowledge point name, grammar structure content, and applicable grammar content corresponding to the grammar rule nodes in the candidate misuse link.

[0125] When the grammar knowledge point name, grammar structure content, and applicable grammar content corresponding to the answer rule label are all consistent with the grammar knowledge point name, grammar structure content, and applicable grammar content corresponding to the grammar rule node in the candidate misuse link, the rule consistency state between the candidate misuse link and the examination rule label is determined to be consistent; when any content is inconsistent, the rule consistency state is determined to be inconsistent.

[0126] Based on the grammatical position, syntactic structure, and grammatical triggering information corresponding to the standard answer in the context of the question, the contextual verification is performed on the answer difference data and misuse performance information associated with the candidate misuse links.

[0127] When the location of the incorrect answer in the answer difference data matches the grammatical location corresponding to the standard answer, the difference type and difference feature in the answer difference data match the syntactic structure information and grammatical trigger information, and the misuse performance information matches the answer deviation content in the rule deviation direction; when any of the incorrect answer location, difference type, difference feature and misuse performance information cannot form a consistent correspondence, the matching state is determined to be a mismatch state.

[0128] Candidate misuse links with both consistent rule status and matching status are identified as valid misuse links; candidate misuse links with inconsistent rule status or mismatching status are not included in the formation of candidate error perception types.

[0129] S5.4: Based on the erroneous cognitive nodes associated with the effective misuse links, classify the erroneous cognitive information corresponding to the erroneous cognitive nodes to form candidate erroneous cognitive types.

[0130] Read valid misuse links; when no valid misuse links exist, mark the current answer-related data as data to be reviewed, and do not participate in the determination of the target error cognition type before review.

[0131] When a valid misuse link exists, read the erroneous cognitive information corresponding to the erroneous cognitive node in the valid misuse link, and associate the erroneous cognitive information, the corresponding syntax rule tag, and the misuse performance information.

[0132] Nodes with consistent erroneous cognitive information, consistent corresponding grammatical rule labels, and consistent misuse behavior information are grouped into the same candidate erroneous cognitive type.

[0133] For each candidate error cognition type, the effective misuse link, error cognition node, misuse performance information and source data identifier are associated and recorded to form a candidate error cognition type.

[0134] S5.5: Based on the direction of the rule deviation, the examination rule label, and the answering rule label, filter historical records of deviations in the same direction that are consistent with the current answering deviation from the historical answering data.

[0135] Based on the direction of rule deviation, the examination rule label, and the answer rule label, the historical answer records corresponding to the same de-identified user identifier are read from the historical answer data and arranged in chronological order of answer time.

[0136] The selection criteria include the direction of rule deviation, the label of the examination rule, and the content of the answer deviation. The corresponding grammar rule labels, answer difference data, and misuse performance information in the historical answer records are matched.

[0137] When the corresponding grammar rule label in the historical answer record is consistent with the examination rule label, the answer difference data is consistent with the rule deviation direction, and the misuse performance information is consistent with the answer deviation content, the historical answer record is identified as a historical same-direction deviation record.

[0138] S5.6: Based on historical same-direction deviation records, match the misuse relation edges corresponding to candidate error cognition types with historical same-direction deviation records, and count the number of occurrences of the same type of deviation, the number of consecutive occurrences, and the number of recurrences after corrective learning to form frequency support values, consecutive support values, and recurrence support values.

[0139] Extract misuse performance information, erroneous cognitive information, response time, and corrective learning records from historical records of similar deviations, and count the total number of historical records of similar deviations.

[0140] For each candidate error cognitive type, the misuse performance information and error cognitive information are matched with the misuse performance information and error cognitive information in the historical same-direction deviation records. The number of historical same-direction deviation records that are successfully matched is counted to obtain the number of times the same type of deviation occurs for the corresponding candidate error cognitive type.

[0141] Read the historical records of successful matching deviations in the order of the answer time, count the maximum number of consecutive occurrences of the same candidate cognitive error type, and obtain the number of consecutive occurrences of the corresponding candidate cognitive error type.

[0142] Read the correction learning record from the historical same-direction deviation record that was successfully matched, count the number of times the same candidate error cognitive type reappeared after the correction learning record was completed, and obtain the number of times the corresponding candidate error cognitive type was reproduced after correction learning.

[0143] Based on the total number of historical records of similar deviations, the number of occurrences of the same type of deviation, the number of consecutive occurrences, and the number of recurrences after corrective learning, the frequency support value, the consecutive support value, and the recurrence support value are calculated respectively, and expressed as follows: ; In the formula, For the first Frequency support values ​​for each candidate cognitive error type For the first Continuous support values ​​for each candidate error cognitive type For the first Support values ​​for the recurrence of candidate cognitive error types, For the first The number of times the same type of bias occurs for each candidate cognitive error type. This represents the total number of historical records of same-direction deviation. For the first The number of consecutive occurrences corresponding to each candidate cognitive error type For the first The number of times each candidate cognitive error type is reproduced after corrective learning. The index of the candidate cognitive error type. This is a set of historical records of deviations in the same direction.

[0144] when When historical records of similar deviations are insufficient to form a historical support assessment, the frequency support value, continuous support value, and recurring support value are all set to 0; when >0 and When =0, both the continuous support value and the recurring support value are set to 0.

[0145] S5.7: Based on frequency support value, continuous support value and recurrence support value, evaluate the confidence of candidate error cognition types and obtain the error cognition confidence value corresponding to each candidate error cognition type.

[0146] Read the frequency support value, continuous support value, and recurrence support value corresponding to each candidate error cognitive type, and use the frequency support value, continuous support value, and recurrence support value as the confidence evaluation data for the corresponding candidate error cognitive type.

[0147] The mean values ​​of frequency support, continuity support, and recurrence support in the confidence evaluation data are calculated to obtain the confidence level of the corresponding candidate false cognitive type, which is expressed as: ; In the formula, For the first The confidence level of each candidate error cognition type.

[0148] By associating the confidence level of the erroneous cognition with the corresponding candidate erroneous cognition type, the confidence level of the erroneous cognition corresponding to each candidate erroneous cognition type is obtained.

[0149] S5.8: Sort candidate error cognition types according to error cognition confidence. When error cognition confidence is the same, sort them in the following order: recurrence support value, continuous support value, frequency support value, and the order in which candidate error cognition types are generated. The candidate error cognition type ranked first is determined as the target error cognition type.

[0150] Read the error cognition confidence level corresponding to each candidate error cognition type, and sort the candidate error cognition types in descending order of error cognition confidence level to form the first sorting sequence.

[0151] When there are candidate false cognitive types with the same false cognitive confidence in the first sorting sequence, the candidate false cognitive types with the same false cognitive confidence are sorted again in descending order of the reproducibility support value to form a second sorting sequence.

[0152] When there are candidate erroneous cognitive types with the same recurrence support value in the second sorting sequence, the candidate erroneous cognitive types with the same recurrence support value are sorted again in descending order of continuous support value to form a third sorting sequence.

[0153] When there are candidate error cognitive types with consecutive support values ​​in the third sorting sequence, the candidate error cognitive types with consecutive support values ​​with consecutive support values ​​are sorted again in descending order of frequency support value to form the fourth sorting sequence.

[0154] When there are candidate error cognitive types with the same frequency support value in the fourth sorting sequence, they are sorted again according to the order in which they were generated to form the target sorting sequence.

[0155] The candidate misperception type that ranks first in the target sorting sequence is determined as the target misperception type.

[0156] S6: Based on the target error cognitive type, perform error correction path reasoning along the error correction relation edges in the English grammar knowledge graph to obtain candidate error correction paths, evaluate the path cost, determine the target error correction path, and generate an English grammar learning path.

[0157] S6.1: Based on the error cognitive information in the target error cognitive type, determine the matching error cognitive node in the English grammar knowledge graph and use it as the starting point for the correction path retrieval.

[0158] Read the error cognitive information and corresponding grammar rule tags from the target error cognitive type, and search for error cognitive nodes in the English grammar knowledge graph whose node names match the error cognitive information and whose corresponding grammar rule tags match. When an error cognitive node is found, it is determined as the starting point for the correction path retrieval. When multiple error cognitive nodes are found, the starting point for the correction path retrieval is determined according to the number of valid misuse links associated with each error cognitive node, from most to least. When no error cognitive node is found, the currently answered data is marked as data to be reviewed and will not be included in the determination of the target correction path before review.

[0159] Starting from the starting point of the correction path retrieval, the system searches for correction relationship edges in the English grammar knowledge graph that start from the correction path retrieval point and reads the correction learning nodes associated with the correction relationship edges. When no correction relationship edge is found, the current answer-related data is marked as data to be reviewed and will not participate in the determination of the target correction path before review.

[0160] The starting point of the correction path retrieval, the correction relation edge, and the correction learning node are associated in the order of connection to form an initial correction path; when the same correction path retrieval starting point corresponds to multiple correction relation edges, the corresponding initial correction paths are formed respectively.

[0161] S6.2: Starting from the starting point of the correction path retrieval, retrieve the associated correction learning nodes along the correction relationship edge, and verify the retrieved path by combining the examination rule label, answer rule label and rule deviation direction to obtain candidate correction paths.

[0162] Read the erroneous cognitive nodes, correction relationship edges, correction learning nodes, correction rule information, and learning content information in the initial correction path, and read the examination rule label, answer rule label, and rule deviation direction.

[0163] Based on the grammar rule tags associated with the correction learning nodes, the consistency of the initial correction path with the examination rule tags and the answer rule tags is checked; when the grammar rule tags associated with the correction learning nodes are consistent with the examination rule tags and the answer rule tags respectively, the initial correction path is determined as a rule tag consistent path.

[0164] Based on the grammatical structure and applicable content corresponding to the learning content information, the consistency of the rule tag consistency path and the grammatical triggering information and deviation content in the rule deviation direction is checked. When the grammatical structure and applicable content corresponding to the learning content information are consistent with the grammatical triggering information and deviation content in the rule deviation direction, the rule tag consistency path is determined as the rule deviation direction consistency path.

[0165] The path with consistent rule deviation direction is recorded in the order of connection of erroneous cognition node, correction relation edge and correction learning node to obtain candidate correction path; when there is no path with consistent rule deviation direction, the current answer related data is marked as data to be reviewed and will not participate in path cost evaluation before review.

[0166] S6.3: Match and verify the candidate correction path with the target error cognition type, rule deviation direction, examination rule label and answer rule label to form candidate correction path verification information.

[0167] Read the erroneous cognitive nodes, correction relationship edges, correction learning nodes, correction rule information, and learning content information in the candidate correction path.

[0168] The erroneous cognitive nodes in the candidate correction path are matched and verified with the target erroneous cognitive type; when the erroneous cognitive information corresponding to the erroneous cognitive node is consistent with the erroneous cognitive information in the target erroneous cognitive type, an erroneous cognitive matching tag is generated.

[0169] The correction rule information in the candidate correction path is matched and verified with the rule deviation direction; when the grammatical structure and applicable grammatical content corresponding to the correction rule information are consistent with the grammatical trigger information and deviation content in the rule deviation direction, a rule deviation direction matching mark is generated.

[0170] The learning content information in the candidate correction path is matched and verified with the examination rule label and the answer rule label respectively; when the grammar rule label corresponding to the learning content information is consistent with both the examination rule label and the answer rule label, a rule label matching mark is generated.

[0171] The error perception matching marker, rule deviation direction matching marker, and rule label matching marker are associated to form candidate correction path verification information; candidate correction paths with complete matching markers enter the path cost evaluation, while candidate correction paths with incomplete matching markers do not participate in the path cost evaluation.

[0172] S6.4: Based on the candidate correction path verification information, and according to the correction learning nodes and historical answer data in the candidate correction path, evaluate the path cost of the candidate correction path to form the candidate correction path cost value.

[0173] Read the candidate correction paths that have passed the verification in the candidate correction path verification information, and perform a non-empty verification on the candidate correction path set; when the candidate correction path set is empty, mark the current answer-related data as data to be reviewed, and do not perform path cost evaluation before review.

[0174] When the candidate correction path set is not empty, count the number of correction learning nodes in each candidate correction path, and determine the maximum number of correction learning nodes in all candidate correction paths as the maximum number of nodes.

[0175] The correction learning nodes in each candidate correction path are matched with the correction learning records in the historical answer data, and the number of times the correction learning records of the corresponding candidate correction path are matched is counted.

[0176] Read the historical same-direction deviation records after the completion of the correction learning from the successfully matched correction learning records, count the number of times the same type of rule deviation reappears after the completion of the correction learning, and obtain the number of times the same type of rule deviation recurs for the corresponding candidate correction path.

[0177] Based on the number of correction learning nodes and the maximum number of nodes, the learning burden value of the corresponding candidate correction path is calculated; based on the number of times the deviation of the same rule is reproduced and the number of times the correction learning record is matched, the reproduction risk value of the corresponding candidate correction path is calculated, expressed as: ; In the formula, For the first The learning burden value of each candidate correction path. For the first The reproducibility risk value of each candidate correction path. For the first The number of correction learning nodes in the candidate correction path. This represents the maximum number of learning nodes among all candidate correction paths. For the first The number of times the deviation of the same rule is reproduced for each candidate correction path. For the first The number of times the correction learning record is matched for each candidate correction path. This is the sequence number of the candidate correction path.

[0178] When the number of corrective learning record matching times is 0, the reproduction risk value lacks historical reproduction statistics, so the reproduction risk value is set to 0.

[0179] The cost of candidate correction paths is calculated based on the learning burden value and the recurrence risk value, and is expressed as follows: ; In the formula, For the first The value of candidate correction paths among candidate correction paths.

[0180] The candidate correction path cost is associated with the corresponding candidate correction path to form the candidate correction path cost.

[0181] S6.5: Sort the candidate correction paths according to their cost value, and determine the target correction path from the candidate correction paths in combination with the direction of rule deviation, thereby generating an English grammar learning path.

[0182] Read the candidate correction path cost value corresponding to each candidate correction path, and sort the candidate correction paths in ascending order of candidate correction path cost value.

[0183] When there are candidate correction paths with the same cost value, read the rule deviation direction matching mark between the candidate correction path and the rule deviation direction, and count the number of grammatical structure contents and the number of grammatically applicable contents that are consistent with the rule deviation direction in the candidate correction path. Sort the candidate correction path with the larger sum of the two quantities first.

[0184] When the sum of the two quantities is the same, read the number of correction learning nodes in the candidate correction path and sort the candidate correction path with fewer correction learning nodes first.

[0185] When the number of correction learning nodes is the same, the candidate correction paths are sorted according to the order in which they were generated, and the candidate correction path ranked first is determined as the target correction path.

[0186] An English grammar learning path is generated in the following order: target error cognition type, correction rule information in the target correction path, correction learning nodes, learning content information, and answer rule tags.

[0187] In summary, this invention determines the target error cognitive type based on the rule deviation direction in the three-label rule deviation state vector. This transforms learner response deviations into cognitively oriented error causal expressions, establishing a stable connection between rule deviation representation and error causal location. The target error cognitive type can summarize and reflect the deviation relationship between assessment rules, response rules, and answer rules, improving the clarity of learner weakness identification and enhancing the consistency and reliability of learning content matching criteria. 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 method for generating English grammar learning paths based on knowledge graph reasoning, characterized in that, include: Acquire relevant data on English grammar learning and preprocess it to obtain basic data for grammar learning; Based on the basic data of grammar learning, we identify grammar rule nodes, error cognition nodes, and correction learning nodes, construct misuse relationship edges between grammar rule nodes and error cognition nodes, and correction relationship edges between error cognition nodes and correction learning nodes, thus forming an English grammar knowledge graph. Acquire grammar question data, current answer data, and historical answer data. Combine the grammar question data and current answer data to identify examination rule tags, answer rule tags, and answer rule tags. Perform contextual analysis on the grammar question data to obtain question context information. We perform rule deviation analysis on the examination rule labels, answer rule labels, and answer rule labels, and combine this with the contextual information of the question to generate a three-label rule deviation state vector that includes the direction of rule deviation. Based on the rule deviation direction in the three-label rule deviation state vector, error cognition retrieval and reasoning are performed along the misuse relation edge in the English grammar knowledge graph to form candidate error cognition types. Then, error cognition confidence is evaluated in combination with historical answer data to determine the target error cognition type. Based on the target error cognitive type, the correction path reasoning is performed along the correction relation edges in the English grammar knowledge graph to obtain candidate correction paths. The path cost is then evaluated to determine the target correction path and generate an English grammar learning path.

2. The method for generating English grammar learning paths based on knowledge graph reasoning as described in claim 1, characterized in that, The specific steps to obtain the basic data for grammar learning are as follows: Acquire English grammar rule texts, grammar question data, question analysis, learning records, and incorrect question records, and perform data cleaning and format standardization to obtain standardized grammar learning data; By analyzing grammatical elements and organizing correction-related information from standardized grammar learning data, basic grammar learning data is obtained.

3. The method for generating English grammar learning paths based on knowledge graph reasoning as described in claim 1, characterized in that, The specific steps for creating an English grammar knowledge map are as follows: Extract grammar rule labels, misuse information, erroneous cognition information, correction rule information, and learning content information from basic grammar learning data, and perform synonym merging and category normalization to form grammar rule nodes, erroneous cognition nodes, and correction learning nodes. Based on the correspondence between grammar rule labels, misuse performance information, and incorrect cognitive information, misuse triples are formed, and misuse relationship edges are constructed between grammar rule nodes and incorrect cognitive nodes; Based on the correspondence between erroneous cognitive information, correction rule information, and learning content information, a correction triplet is formed, and a correction relationship edge is constructed between erroneous cognitive nodes and correction learning nodes. By associating and storing grammar rule nodes, error cognition nodes, correction learning nodes, misuse relation edges, and correction relation edges, an English grammar knowledge graph is formed.

4. The method for generating English grammar learning paths based on knowledge graph reasoning as described in claim 1, characterized in that, The process of obtaining the examination rule label, the answer rule label, and the answer rule label refers to acquiring grammar question data and current answer data, extracting the question stem, question analysis, and standard answer from the grammar question data, and extracting the user answer from the current answer data; identifying the examination rule label based on the question stem and question analysis, and placing the user answer and standard answer into the corresponding positions in the question stem to perform grammar rule recognition, thereby obtaining the answer rule label and the answer rule label.

5. The method for generating English grammar learning paths based on knowledge graph reasoning as described in claim 1, characterized in that, The aforementioned acquisition of the question context information refers to performing context analysis on the question stem, question analysis, and standard answer, extracting syntactic structure information, grammatical trigger information, and the grammatical position of the standard answer, and associating the syntactic structure information, grammatical trigger information, and the grammatical position of the standard answer to form the question context information.

6. The method for generating English grammar learning paths based on knowledge graph reasoning as described in claim 1, characterized in that, The specific steps for generating the three-label rule deviation state vector containing the rule deviation direction are as follows: Using the examination rule tags as rule references, the answer rule tags and the answer rule tags are compared with the examination rule tags in terms of rule category and grammatical function to form a three-tag rule reference relationship; Based on the three-label rule reference relationship, the deviation status of the answer rule label relative to the examination rule label and the matching status of the answer rule label relative to the examination rule label are verified to form the three-label rule deviation relationship; Based on the three-label rule deviation relationship, the difference between the answer rule label and the answer rule label is mapped to the syntactic structure information and grammatical trigger information in the question context information to form the rule deviation direction; Based on the direction of the rule deviation, the relationship between the three-label rule deviations and the contextual information of the question are combined and encoded to generate a three-label rule deviation state vector containing the direction of the rule deviations.

7. The method for generating English grammar learning paths based on knowledge graph reasoning as described in claim 1, characterized in that, The specific steps for forming candidate error cognitive types are as follows: Read the rule deviation direction from the three-label rule deviation state vector, and match the rule deviation direction with the examination rule label and the answer rule label to form the misuse link retrieval conditions; Based on the retrieval criteria for misuse links, misuse links associated with the examination rule tags and answer rule tags are retrieved along the misuse relationship edges in the English grammar knowledge graph to form candidate misuse links; Based on the answer rule tags, the rule pointers corresponding to the candidate misuse links are checked to determine the rule consistency status between the candidate misuse links and the examination rule tags. Based on the question context information, the matching status between the candidate misuse links and the question context information is checked to form a valid misuse link. Based on the erroneous cognitive nodes associated with the effective misuse links, the erroneous cognitive information corresponding to the erroneous cognitive nodes is classified to form candidate erroneous cognitive types.

8. The method for generating English grammar learning paths based on knowledge graph reasoning as described in claim 1, characterized in that, The specific steps for determining the target's cognitive error type are as follows: Based on the direction of rule deviation, the examination rule label, and the answer rule label, filter historical records of deviations in the same direction that are consistent with the current answer deviation from the historical answer data; Based on historical records of similar deviations, the misuse relation edges corresponding to candidate error cognitive types are matched with historical records of similar deviations. The number of occurrences of similar deviations, the number of consecutive occurrences, and the number of recurrences after corrective learning are counted to form frequency support values, consecutive support values, and recurrence support values. Based on frequency support values, continuity support values, and recurrence support values, the confidence level of candidate error cognition types is evaluated to obtain the error cognition confidence level corresponding to each candidate error cognition type; Candidate error cognitive types are sorted according to their error cognitive confidence. When the error cognitive confidence is the same, they are sorted in the following order: recurrence support value, continuous support value, frequency support value, and the order in which the candidate error cognitive types were generated. The candidate error cognitive type ranked first is determined as the target error cognitive type.

9. The method for generating English grammar learning paths based on knowledge graph reasoning as described in claim 1, characterized in that, The specific steps for obtaining the candidate correction path are as follows: Based on the error cognitive information in the target error cognitive type, the matching error cognitive node is determined in the English grammar knowledge graph and used as the starting point for the correction path retrieval; Starting from the starting point of the correction path retrieval, the associated correction learning nodes are retrieved along the correction relationship edges. The retrieved path is then verified by combining the examination rule label, answer rule label, and rule deviation direction to obtain candidate correction paths.

10. The method for generating English grammar learning paths based on knowledge graph reasoning as described in claim 1, characterized in that, The specific steps for generating the English grammar learning path are as follows: The candidate correction paths are matched and verified with the target error cognition type, rule deviation direction, examination rule label and answer rule label to form candidate correction path verification information; Based on the verification information of the candidate correction path, the path cost of the candidate correction path is evaluated according to the correction learning nodes and historical answer data in the candidate correction path, and the cost value of the candidate correction path is formed. Candidate correction paths are sorted according to their cost value, and the target correction path is determined from the candidate correction paths based on the direction of rule deviation, thus generating an English grammar learning path.

Images