Teaching course recommendation method and system based on English learning data

Abstract

The invention discloses a teaching course recommendation method and system based on English learning data, particularly relates to the field of semantic processing, is used for solving the problem of poor pertinence of a traditional English learning course, and comprises the following steps: aiming at systematic difference of native language and English expression on a syntactic structure, constructing a language structure difference map and extracting a structure offset label; the method comprises the following steps: performing dependency syntactic analysis on semantic consistent sentence pairs of multi-language aligned corpora to generate a structural difference feature set; on the basis, a semantic-grammar dimension mapping graph is constructed in a classified mode, and a general structure expression vector is established for graph nodes. A teaching course is divided into knowledge point units in combination with a context label, and a mapping relation between a structure label and the course is established. The system performs structure analysis and semantic matching on sentences input by the learner, identifies structure migration type expression errors, and recommends accurate teaching content according to context and structure labels.

Description

Technical Field

[0001] The present invention relates to the field of semantic processing technology, and more particularly, to a teaching course recommendation method and system based on English learning data. Background Art

[0002] In the process of learning English, especially at the beginner and intermediate levels for non-native speakers, learners frequently exhibit a phenomenon in which they superficially conform to grammatical rules, but deviate significantly from native speakers' expressions in terms of logic, word order, or semantic details. This phenomenon stems from language transfer interference, where learners unconsciously borrow syntactic, semantic, or pragmatic structures from their native language when producing English expressions, resulting in a stable pattern of errors that are structurally legal but unnatural in expression. For native Chinese learners, for example, common examples of such interference include omitting the subject, misusing verb tenses, directly replacing the passive voice or emphatic structures of English with linear subject-verb-object structures, and misapplying Chinese logical order or using uncoordinated expressions when expressing complex logic such as cause and effect, concession, and comparison. Traditional English teaching systems often misclassify these errors as "grammatical errors" or "inadequate expression," overlooking the fact that these errors may be the result of a stable and systematic process of native language transfer.

[0003] Therefore, relying solely on vocabulary and grammar mastery is insufficient to support truly personalized course recommendations. Addressing this issue requires building a mechanism that can reverse engineer the learner's native language structural intervention pathways based on their historical language generation data, identifying the types of expression structures where transfer barriers exist. Based on this, it can recommend instructional content with structural reconstruction capabilities, shifting from error correction to course recommendations focused on correcting transfer interference. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a teaching course recommendation method based on English learning data to solve the problems raised in the above-mentioned background technology.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A teaching course recommendation method based on English learning data includes the following steps:

[0007] S1: Extract equivalent semantic expressions from native language and English corpora, perform dependency parsing and structural component extraction, and generate a set of cross-language structural difference features;

[0008] S2: Classify the collection of structurally different segments into multi-dimensional grammatical types and construct a semantic-grammatical dimension mapping map;

[0009] S3: Establish a universal expression template for each connection node in the graph to generate a searchable structure vector;

[0010] S4: Divide the teaching course into knowledge point units according to the context scenario, and establish a knowledge point-level mapping relationship between the course content and the graph nodes through structural offset labels;

[0011] S5: Generate a structural vector of the user input sentence based on syntactic parsing and semantic extraction, and compare it with the structural vector of the corresponding universal expression template in the graph to identify structural transfer errors.

[0012] S6: Recommend teaching courses based on the context of the user's input sentence and the structural shift labels corresponding to structural transfer expression errors.

[0013] In a preferred embodiment, in S1, equivalent semantic expressions are extracted from the native language and English corpus, dependency parsing and structural component extraction are performed, and a set of cross-language structural difference features is generated, specifically including:

[0014] Perform semantic parsing on the aligned corpus of native language and English, and select sentence pairs with consistent meanings;

[0015] Dependency structure analysis is performed on the native language expression and English expression in the sentence pairs with consistent sentence meaning, and a dependency structure tree is generated based on the inter-word dependency relationship;

[0016] The syntactic backbone chain, intra-sentence modification relations and subordinate structure boundaries are extracted from the dependency structure tree, a set of syntactic structure components is established, and a syntactic function position alignment operation is performed on the set of structure components. The grammatical fragments with structural deviations in the expression are marked as structural difference fragments, and a set of language structure difference fragments is generated.

[0017] In a preferred embodiment, in S2, the set of structural difference segments is classified into multi-dimensional grammatical types, and the construction of a semantic-grammatical dimension mapping map specifically includes:

[0018] Preset multi-dimensional grammatical structure offset tags, and divide the structural difference segments in the structural difference segment set into dimensions according to the structural offset tags to form offset segment groups;

[0019] Semantic extraction is performed on the offset fragment group to form a joint semantic space, and an association mapping between the joint semantic space and the structural offset label is constructed. The structural offset label is used as the graph node and the joint semantic space is used as the connecting node to establish a semantic-syntactic dimension mapping graph.

[0020] In a preferred embodiment, in S3, establishing a universal expression template for each connection node in the graph to generate a searchable structure vector specifically includes:

[0021] In the semantic-syntactic dimension mapping graph, semantic item overlap analysis is performed on the joint semantic space corresponding to all connected nodes, and semantic items existing in multiple structural offset labels are integrated to construct a semantic intersection set.

[0022] Using the semantic items in the semantic intersection set as index keys, aggregate all corresponding structural offset labels to generate a structural offset label group;

[0023] According to the structure offset label group, the semantic items of the semantic intersection set are rewritten into a universal expression template based on the structure offset label and converted into a searchable structure vector.

[0024] In a preferred embodiment, rewriting the semantic items of the semantic intersection set into a universal expression template based on a structure offset tag and converting it into a searchable structure vector specifically includes:

[0025] Obtain semantic items with the same structural offset tag group, obtain the syntactic backbone chain of the semantic items corresponding to the structural difference segments, and extract the corresponding word class arrangement rules of the backbone verbs, subjects, and core objects;

[0026] According to the grammatical change mode specified by the current structure offset tag group, abstract placeholders are sequentially inserted into the generated syntactic backbone chain word order arrangement rules to form a general expression template grammar chain with an offset structure;

[0027] The structural expression vector is constructed according to the three-dimensional factors of word class arrangement, structural offset label position, and label combination corresponding to the general expression template grammatical chain.

[0028] In a preferred embodiment, in S4, the teaching course is divided into knowledge point units according to the context scenario, and the knowledge point level mapping relationship between the course content and the graph nodes is established through the structure offset label, specifically including:

[0029] Classify teaching courses according to actual communication context scenarios, label each course with the communication context label, and extract keywords for the knowledge point units of the teaching courses;

[0030] Based on the grammatical structure offset label keywords corresponding to the graph nodes, the relationship mapping between the structure offset labels and the teaching course knowledge point units under different context labels is carried out through keyword matching, and a course recommendation decision engine based on the structure offset label is established according to the relationship mapping.

[0031] In a preferred embodiment, in S5, a structural vector of the user input sentence is generated based on syntactic parsing and semantic extraction, and a vector comparison is performed with the structural vector of the corresponding universal expression template in the graph. Identifying structural transfer errors specifically includes:

[0032] Perform word tagging and word dependency structure analysis on user input sentences based on grammatical structure offset labels, generate a dependency structure tree based on inter-word dependency relationships, and extract the syntactic backbone chain;

[0033] Combine the syntactic backbone chain with the structural offset labeling of the word item to construct the structural vector of the input sentence;

[0034] Perform semantic extraction on user input sentences, map the extraction results to the joint semantic space corresponding to the connected nodes in the semantic-syntactic dimension mapping graph, and perform semantic matching retrieval;

[0035] If no matching semantic item is found, the user input statement is directly marked as grammatically incorrect;

[0036] If a matching semantic item is successfully retrieved, the corresponding universal expression template structure vector is obtained and compared with the structure vector of the input sentence to identify structural transfer expression errors.

[0037] In a preferred embodiment, the vector comparison with the structural vector of the input sentence and the identification of structural transfer type expression errors are specifically performed by comparing the parameters in the vector item by item. If there are missing vector parameters or inconsistent orders, the input sentence is identified as a structural transfer type expression error.

[0038] In a preferred embodiment, in S6, recommending a teaching course based on the context of the user input sentence and the structural shift label corresponding to the structural transfer error specifically includes:

[0039] When a user input sentence is identified as a structural shift error, the context label corresponding to the semantic usage scenario of the input sentence is extracted and combined with the structural shift label combination in the general expression template as a course screening condition;

[0040] The screening conditions are input into the course recommendation decision engine, and the final teaching course recommendation results are output.

[0041] On the other hand, the present invention provides a teaching course recommendation system based on English learning data, including a structure extraction module, a graph construction module, a template generation module, an error recognition module and a course recommendation module:

[0042] Structure Extraction Module: This module performs dependency structure analysis on equivalent sentence pairs between the native language and English, extracts syntactic backbone chains, modification relations, and subordinate boundaries, and marks structural deviation segments according to component alignment to form language structure difference segments.

[0043] Graph construction module: divides the structural difference segments into offset segment groups, extracts the semantic items within the groups to construct a joint semantic space, and establishes a semantic-syntactic dimension mapping graph with offset labels as nodes;

[0044] Template generation module: Extracts the corresponding syntactic backbone chain sequence based on the structural offset tag group, embeds abstract placeholders to generate a universal expression template grammar chain, and generates a structural expression vector based on the word class sequence and tag combination;

[0045] Error Identification Module: This module constructs structural vectors and extracts semantic terms from user sentences, searches the joint semantic space in the graph, and matches the common template vector. If parameters are missing or the order is shifted, it identifies a structural transfer error.

[0046] Course recommendation module: collects input sentence context labels, combines them with error-corresponding structural offset label groups to form screening conditions, drives the course recommendation decision engine to match graph nodes with associated courses, and generates a target course recommendation list.

[0047] The technical effects and advantages of the present invention's teaching course recommendation method and system based on English learning data are as follows:

[0048] In response to the expression deviations that occur in learners during the process of language structure transfer, an intelligent recommendation mechanism that integrates semantic analysis and syntactic structure mapping is constructed. This method extracts semantically equivalent but structurally different expressions from the native language and English corpus, constructs a set of structural difference features, and deeply reveals the transfer characteristics of the two languages ​​at the syntactic level. Structural differences are further classified through multi-dimensional grammatical offset labels, and a mapping map is established with the corresponding semantic items to form a clear correspondence between semantics and structure. On this basis, a quantifiable structural expression vector is abstractly generated to effectively support subsequent rapid matching and structural consistency judgment. At the same time, the teaching course content is finely divided according to context and structural knowledge points, and mapped with grammatical structure labels to achieve structural-level knowledge point binding.

[0049] After identifying structural transfer errors in the expressions input by learners, the system can combine semantic context and structural labels to accurately match teaching courses suitable for the current learning status, forming a complete closed-loop diagnosis-identification-recommendation link, greatly improving the targeted nature of course recommendations and the effectiveness of teaching interventions. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 This is a schematic diagram of a teaching course recommendation method based on English learning data according to the present invention;

[0051] Figure 2 This is a structural diagram of a teaching course recommendation system based on English learning data according to the present invention. DETAILED DESCRIPTION

[0052] The following will provide a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0053] Example 1

[0054] Figure 1 The present invention provides a teaching course recommendation method based on English learning data, which includes the following steps:

[0055] S1: Extract equivalent semantic expressions from native language and English corpora, perform dependency parsing and structural component extraction, and generate a set of cross-language structural difference features;

[0056] S2: Classify the collection of structurally different segments into multi-dimensional grammatical types and construct a semantic-grammatical dimension mapping map;

[0057] S3: Establish a universal expression template for each connection node in the graph to generate a searchable structure vector;

[0058] S4: Divide the teaching course into knowledge point units according to the context scenario, and establish a knowledge point-level mapping relationship between the course content and the graph nodes through structural offset labels;

[0059] S5: Generate a structural vector of the user input sentence based on syntactic parsing and semantic extraction, and compare it with the structural vector of the corresponding universal expression template in the graph to identify structural transfer errors.

[0060] S6: Recommend teaching courses based on the context of the user's input sentence and the structural shift labels corresponding to structural transfer expression errors.

[0061] In S1, equivalent semantic expressions are extracted from the native language and English corpus, dependency syntactic analysis and structural component extraction are performed, and a set of cross-language structural difference features is generated.

[0062] Establish a cross-language corpus resource to build an expression alignment relationship between the mother tongue (for non-native English learners) and English. The corpus resources are selected from national standardized test corpora and bilingual control sentences in mainstream textbooks. In the specific implementation, a semantic parsing engine built based on a role semantic framework is introduced to perform semantic role labeling on the mother tongue and English sentences respectively, extracting the core semantic frame of the verb and its corresponding semantic argument structure (such as agent, patient, tool, etc.). By comparing the verb type, the number and category of argument roles, and the matching of corresponding semantic content in the main semantic frames of the two language sentences, it is determined whether they belong to a sentence pair with consistent semantic attribution. If the two have the same semantic framework, it means that the number of core arguments and semantic roles are completely aligned, and dependency structure parsing operations are performed on the mother tongue expression and the corresponding English expression respectively. The goal of dependency structure parsing is to construct a syntactic dependency tree with verbs or syntactic core words as the root node, and clarify the dependency and modification relationships between the words in the sentence. The dependency structure of native language sentences is constructed using a syntactic parsing model based on language characteristics. For example, for Chinese expressions, a dependency model trained on hierarchical phrase grammar features is used to accurately identify core components such as subject-verb relationships, attributives, adverbials, complements, and prepositional structures. For English sentences, dependency relationship identification rules trained specifically on object-complement structures, auxiliary tense chains, and prepositional phrases are used to analyze the dependency paths between the main verb, its arguments, and its modifiers. After parsing, the dependency structure of each sentence is represented as a tree data structure, where each node represents a lexical item and each edge represents a subordinate relationship. For example, the sentence "She gave him a gift" forms a dependency tree with "gave" as the root, "she" as the subject, "him" as the indirect object, and "gift" as the direct object.

[0063] After the dependency structure tree is established, core structural information is extracted from it. First, the syntactic backbone chain of each dependency tree is identified. This is the core semantic path extending outward from the root node (usually the main verb), which mainly includes the basic sentence skeleton composed of subject, predicate, and object. For example, in English, the basic chain structure is "subject-predicate-object", while in Chinese it may be expressed as a "topic-rheme" structure. Secondly, the modification relationships existing in the sentence are extracted, such as adjectives modifying nouns, adverbs modifying verbs, and prepositional phrases modifying verbs or nouns. The attachment path between each type of modifier and its modified object is recorded. All subordinate structure boundary positions are identified, such as the start and end positions of attributive clauses, object clauses, and non-restrictive clauses, as well as their attachment points to the main structure. The above three types of elements are uniformly encoded to form a set of syntactic structure components. This set can represent the hierarchical relationship, positional attributes, and functional labels of each component in the sentence. For example, in the sentence “The book that he gave me was interesting”, the set of structural components should clearly identify “that he gave me” as an attributive clause, which is attached to “book”, and mark its syntactic function as modification.

[0064] After obtaining a set of structural components for Chinese and English sentences, a functional alignment process is performed between the two languages ​​based on the extracted syntactic functions and the position of each corresponding lexical item. This alignment considers the matching of lexical item function labels (e.g., subject to subject, object to object) and their actual positional order within the sentence to determine whether there is significant structural shift. Types of shift primarily include adjustments to word order (e.g., postpositioned structures in Chinese versus prepositioned structures in English), changes in the position of modifying components (e.g., postpositioned attributives to prepositioned attributives), and changes in the nesting level of subordinate structures. When a functional component exhibits significant inconsistency in position or structural nesting between English and Chinese expressions, it is identified as a structural shift segment. All grammatical segments identified as structural shifts are uniformly annotated and constitute a set of linguistic structural difference segments. Each entry in this set records the language type, structural shift type, corresponding syntactic function label, sentence position index, and original dependency path, forming a structural resource library for subsequent graph construction and template abstraction.

[0065] In S2, the collection of structurally different fragments is classified into multi-dimensional grammatical types to construct a semantic-grammatical dimension mapping map.

[0066] The preset multi-dimensional syntactic structure offset label system should cover the common offset types in language structure changes, including at least but not limited to seven basic dimensions such as "modifier position offset", "subject default", "syntactic function replacement", "subordinate structure extension", "word order reconstruction", "verb structure delay", and "recovery of non-overt subject". In the specific design, for each type of offset dimension, it is necessary to clarify the applicable dependency structure manifestation form, triggering syntactic environment, and recognizable attachment path pattern. Taking "modifier position offset" as an example, it is set that this label is applied to the structural difference segment that changes from "noun pre-modification" to "noun post-modification". For example, "the man with a hat" in English corresponds to the Chinese expression "戴帽子的男人", where the position of the modifier is replaced in the syntactic order. The dimension division of the offset label takes the set of structural difference segments as the input, compares the structural offset features shown by each segment item by item, and classifies them according to the label definition rules. In the classification process, it not only depends on the morphological and syntactic position indexes, but also combines the directional features indicating the master-slave relationship in the dependency structure path to ensure the grammatical rigor of the label division. Each segment classified into the same label dimension will be identified as the same "offset segment group", that is, the set of grammatical segments with consistent offset features in terms of structural performance. For example, the segments that appear in multiple sentence pairs and show that "the attributive clause modifies the noun post-positionally in English, while the pre-modification is used in Chinese" will all be classified into the segment group under the "post-position of attributive" offset label. Each formed offset segment group has a unique label identifier and a set of intra-group structural difference instances.

[0067] Conduct semantic attribution analysis on the sentence segments in each offset segment group. Taking the segments "with a hat" (English) and "戴帽子的" (Chinese) as an example, both can be classified under the semantic intention of "attribute modification - appearance feature". When extracting semantics, first identify the key word items that play the main information function in the structural segment. For example, "hat" or "帽子" represents a specific attribute entity, and "with" and "戴" represent the attribute attachment relationship. These key words together form a brief semantic expression unit. The semantic expression unit is recorded in the form of an expression unit combination to form a unified coding method. After completing the extraction of semantic units, aggregate the semantic expressions of all the expression segments from the same offset segment group uniformly to form the semantic set of this label. This set, as the original component of the joint semantic space, identifies the semantic bearing direction behind the current structural offset. When there is a tendency of expression convergence between the semantic clustering sets of different structural offset labels, bind multiple semantic items to the same joint semantic space in the form of offset labels. For example, for the offset segments of the word order inversion type involved in the "post-position of attributive" label, the segment semantics involved in this label can be co-clustered into the same joint semantic space through semantic classification, initially forming a node for the post-position of attributive.

[0068] After the joint semantic space is constructed, a bidirectional mapping relationship is established between it and the structural offset labels. Each offset label must point to the joint semantic space on which it relies, indicating the semantic representation carried by the label change. Each joint semantic space must also point to all dependent structural offset label groups. This mapping relationship is represented by a graph structure: the offset labels serve as graph nodes, the joint semantic spaces serve as connecting nodes, and the directionality of the edges points unidirectionally from structure to semantics. Ultimately, all mapping relationships form a semantic-syntactic dimension mapping graph.

[0069] In S3, a universal expression template is established for each connected node in the graph to generate a searchable structure vector.

[0070] After the semantic-syntactic dimension mapping graph is completed, a mapping between connecting nodes and structural offset labels is established. Semantic item overlap analysis is performed within the joint semantic space bounded by each connecting node in the graph. Semantic item identification criteria rely on the semantic function attribution, core semantic verbs, agent / object entity roles, and the positioning of modifiers within the semantic function space. Equivalence is not determined through literal term matching, but rather through classification and merging using a semantic role encoding system. In practice, the semantic expression fragments corresponding to each structural offset label in the joint semantic space are first traversed, their semantic unit labels (e.g., "attribute attachment," "action target," "spatial orientation") are extracted, and overlapping hits under multiple labels are counted. Semantic items that appear in two or more offset labels are grouped into a "semantic intersection set." Each semantic intersection item uses its semantic identifier code as a unique index key, and its source label set is established through semantic item backtracking logic. Based on this, all structural offset labels associated with each semantic item are systematically aggregated to form a corresponding structural offset label group. For example, the semantic item "subject action + target position" may appear under the three offset labels of "subject advance," "adverbial postposition," and "prepositional structure shift." In this case, the label group for this semantic item is {subject advance, adverbial postposition, prepositional structure shift}. This group no longer preserves the original sentence structure differences, but serves as a collection of syntactic changes that universally express abstraction.

[0071] For each semantic intersection term, a syntactic backbone chain is extracted from all structurally different segments, focusing on the main verb, subject, core object, prepositional structure, and modifier positions. Subsequently, based on the change patterns defined by each structural shift tag, such as "moving the attributive from the front of the noun to the end of the sentence" or "inserting the adverbial after the predicate," these grammatical behaviors are applied one by one to a unified backbone chain construction logic, forming a multi-version structural layout. Specific lexical items are not retained in the structural layout, only the word class placeholders and word order information are retained. For example, "[NP-subj]+[VP-core]+[PP-attr]" and "[NP-subj]+[VP-core]+[NP-attr]+[PP]" are versions of a common expression template grammatical chain under two shift tags. Under the same semantic term, all template grammatical chains are differentially aligned, and the changed structural positions and tag combinations are identified. Based on these structural transformation combinations, the embedding order, shift position, and structural combination index of each word class and structural segment in the template expression are generated. To organize this information into a vector representation, a triple encoding scheme is employed: first, a structural word order code is generated based on word class arrangement; second, offset positions are marked with a Boolean sequence (e.g., "subject postposition" is "0-1"); and third, the structural offset labels are combined and encoded into a position mask vector. These three encodings are combined under a unified template to generate a structural expression vector. The vector dimension is set according to the maximum structural position information in the template, ensuring equal dimensionality for all templates. This structural vector does not carry information about specific linguistic terms but serves only as an index of the abstract sentence representation of the semantic term under multiple structural labels. For example, "The boy ran with a hat" and its English equivalent, "The boy ran with a hat," are semantically equivalent, but differ in their structural representation by "attributive preposition" and "adverbial postposition." Based on this, the semantic term is labeled with the structural offset label group {attributive preposition, adverbial postposition}. After extracting the corresponding syntactic backbone chain, a unified general expression template is abstracted: [NP-subj]+[VP]+[Attr / PP], in which the modifier position is embedded as a variable structural parameter in the template. The template is then converted into a structural vector expression based on the three elements of word order, structural offset position Boolean marker, and offset label combination number. For example, the vector [1|1,0|E21] corresponds to the word order type, offset position combination, and structural offset label encoding ("1" indicates that the word order type number corresponds to "subject + predicate + modifier", "1,0" indicates the position of the modifier before or after, and "E21" is the structural offset label group number for fast indexing and comparison). This achieves a unified quantitative representation and searchable modeling of cross-language structural difference expressions.

[0072] In S4, teaching courses are divided into knowledge point units according to contextual scenarios, and a knowledge point-level mapping relationship is established between course content and graph nodes through structural offset labels.

[0073] In actual teaching courses, different communicative scenarios often determine the functional structure and structural usage tendencies of language expressions. Therefore, to achieve an efficient mapping between course content and structural deviation labels, it is necessary to first perform scenario classification on all pending courses. Specifically, typical contextual information in course content should be classified based on communicative purpose. Scenario categories such as "self-introduction," "asking for directions while shopping," "career planning," "expressing opinions," "conditional assumptions," and "behavioral persuasion" all possess stable pragmatic characteristics and grammatical structure preferences. The course context labeling process begins with a structured extraction of the course's teaching objectives, example sentences, and exercise design content, recording the functional sentence patterns and thematic semantics involved in each course. For each course, context attribution is determined based on its teaching objectives and language activity type, against the context labels set in the scenario library. If multiple contextual components exist in the course content, the context label representing the dominant language activity is prioritized. After completing the context labeling, keywords for the knowledge points and units involved in the course are extracted. Keyword extraction focuses on verb phrases, prepositional phrases, conjunctions, and fixed sentence structures found in course descriptions, example sentences, and exercise instructions. All keywords are uniformly normalized and used as candidates for the course's functional structure keywords. All extracted keywords form a bag of words for the course's knowledge points, which is then used for matching with structural offset tags.

[0074] In the semantic-syntactic dimension mapping graph, each graph node is associated with a structural offset label, which defines a corresponding set of keywords (i.e., grammatical inflection keywords in the abstract expression template). For example, the "attributive postposition" label contains relative clause initiators, and the "adverbial preposition" label contains linking adverbs. The course knowledge point keywords are matched with the keywords under the structural offset labels. If there is keyword overlap or semantic consistency (determined by keyword hyponymy mapping), a mapping relationship between the structural offset label and the course is established, and the contextual label in which it is located is marked. By statistically analyzing and archiving these matching relationships, a basic mapping rule set for the course recommendation decision engine is constructed. The decision engine receives the structural offset label group and the contextual label as input, queries the established label-course mapping table, and outputs a set of courses that meet the current structural training needs and contextual requirements. The decision engine can use a label priority ranking mechanism. When multiple courses are matched simultaneously, the optimal recommendation is selected based on the complexity of the offset label combination or the number of structural labels covered by the course. This process does not rely on the platform or external rule library and is entirely based on the engineering matching rules between structural labels and contextual scenarios.

[0075] In S5, the structural vector of the user input sentence is generated based on syntactic parsing and semantic extraction, and is compared with the structural vector of the corresponding general expression template in the graph to identify structural transfer expression errors.

[0076] The user input sentence is annotated with grammatical structure offset labels and the dependency structure of the terms is parsed. A dependency structure tree is generated based on the inter-word dependency relationship and the syntactic backbone chain is extracted. The syntactic backbone chain is combined with the structural offset label annotation of the terms to construct the structural vector of the input sentence.

[0077] A grammatical structure analysis operation is performed on the original sentence input by the learner. The first task of this operation is to label each word in the sentence with a structural deviation label. Word tagging is not limited to part-of-speech tagging. Instead, it is based on the structural deviation label system established in the aforementioned semantic-syntactic graph to clearly mark structural features in the input sentence, such as word order changes, modification position adjustments, and inserted clauses. After the structural deviation labels of the words are labeled, the dependency structure parsing stage begins. This stage uses the input sentence as a unit and constructs a complete dependency structure tree based on the dependency relationships between words. Dependency relationships can include subject-predicate structures (such as between the subject and the predicate verb), object-verb structures, modification relationships (such as adjectives modifying nouns), and subordinate relationships between complements. Finally, a structural vector of the original sentence input by the learner is constructed using the same process as the general expression template construction and vector conversion method.

[0078] After the structural vector is constructed, a semantic extraction operation is performed on the input sentence. The semantic extraction stage focuses on extracting core semantic units from the sentence, such as the subject's action intention, the object category, the time condition, or the reasoning intention. The extraction results are uniformly mapped to the joint semantic space pointed to by the connection node in the semantic-syntactic dimension mapping graph constructed above, and a semantic item matching retrieval operation is performed. If the semantics expressed by the input sentence is not in the set of known semantic items in the current graph (that is, it fails to hit any joint semantic space item), it can be directly determined that the sentence has an error at the semantic construction level, and can be marked as a "semantic miss-type grammatical error."

[0079] If a matching semantic term is successfully retrieved, the structure vector of the universal expression template to which it is bound can be located. This vector is pre-generated using a standard expression template, ensuring a stable structure and clear labels. At this point, the structure vector of the user's input sentence is compared against the template structure vector at a parameter level. This comparison examines the word class sequence, structural offset label encoding, and word order index to check for omissions, misplacements, or inconsistencies. For example, if the template structure is "[NP-subj]+[VP-core]+[PP-attr]" and the offset label group is numbered "E21," while the corresponding position in the input sentence is "[NP-subj]+[VP-core]," this indicates a missing object. Alternatively, if the offset label sequence is "E20," this indicates a lack of the required sentence structure. Both types of errors are considered structural transfer errors. These errors do not stem from improper semantic construction, but rather from a discrepancy between the grammatical implementation path of the expression and the pre-set structure. Essentially, they result from the influence of native language expression habits when expressing the same meaning, leading to incorrect expressions. These errors require structural adjustment training through instructional courses.

[0080] In S6, teaching courses are recommended based on the context of the user's input sentence and the structural shift labels corresponding to structural transfer expression errors.

[0081] Semantic keywords are extracted from the semantic vector generated in the completed semantic extraction phase. Combined with the contextual information of the input sentence, the language context in which the sentence appears is determined, such as "expressing personal preferences," "making a request," "description," or "expressing a plan." This context label must be consistent with the context categories defined in the curriculum. Simultaneously, the structural shift label groups corresponding to identified structural shift errors are extracted as parallel input conditions. These structural shift label groups were established during the earlier construction of the general expression template and represent specific categories of syntactic shifts. After the two sets of input conditions are determined, the course screening phase begins. This phase relies on the previously established basic mapping rule set to match structural shift label nodes with course content unit nodes. Course content units have been associated and indexed according to three dimensions: knowledge points, context labels, and structural shift labels during the course division phase. Course unit nodes that intersect with the current structural shift label group are searched. If multiple matching nodes exist, the matching degree is used as a ranking criterion, prioritizing courses with high label overlap and similar scenarios. Finally, the candidate course units identified in this screening process are packaged as recommendation results and output to the user interface. The output content includes not only the course title, but also key information such as structural offset label description, context application scope, knowledge point keywords, etc.

[0082] Example 2

[0083] The difference between Example 2 of the present invention and Example 1 is that this example introduces a teaching course recommendation system based on English learning data.

[0084] Figure 2 The present invention provides a structural diagram of a teaching course recommendation system based on English learning data, which includes a structure extraction module, a graph construction module, a template generation module, an error recognition module, and a course recommendation module:

[0085] Structure Extraction Module: This module performs dependency structure analysis on equivalent sentence pairs between the native language and English, extracts syntactic backbone chains, modification relations, and subordinate boundaries, and marks structural deviation segments according to component alignment to form language structure difference segments.

[0086] Graph construction module: divides the structural difference segments into offset segment groups, extracts the semantic items within the groups to construct a joint semantic space, and establishes a semantic-syntactic dimension mapping graph with offset labels as nodes;

[0087] Template generation module: Extracts the corresponding syntactic backbone chain sequence based on the structural offset tag group, embeds abstract placeholders to generate a universal expression template grammar chain, and generates a structural expression vector based on the word class sequence and tag combination;

[0088] Error Identification Module: This module constructs structural vectors and extracts semantic terms from user sentences, searches the joint semantic space in the graph, and matches the common template vector. If parameters are missing or the order is shifted, it identifies a structural transfer error.

[0089] Course recommendation module: collects input sentence context labels, combines them with error-corresponding structural offset label groups to form screening conditions, drives the course recommendation decision engine to match graph nodes with associated courses, and generates a target course recommendation list.

[0090] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters and thresholds in the formulas are set by technicians in this field according to actual conditions.

[0091] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer program are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center via a wired (e.g., infrared, wireless, microwave, etc.) method. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state drive.

[0092] Those skilled in the art will appreciate that the modules and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0093] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and modules described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0094] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the modules is only a logical function division. In actual implementation, there may be other division methods, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or modules, which can be electrical, mechanical or other forms.

[0095] The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical modules, and may be located in one place or distributed across multiple network modules. Some or all of the modules may be selected to achieve the purpose of this embodiment according to actual needs.

[0096] In addition, each functional module in each embodiment of the present application may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0097] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0098] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

[0099] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A teaching course recommendation method based on English learning data, characterized in that: The steps include: S1: Extract equivalent semantic expressions from native language and English corpora, perform dependency parsing and structural component extraction, and generate a set of cross-language structural difference features; S2: Classify the collection of structurally different segments into multi-dimensional grammatical types and construct a semantic-grammatical dimension mapping map; S3: Establish a universal expression template for each connection node in the graph to generate a searchable structure vector; S4: Divide the teaching course into knowledge point units according to the context scenario, and establish a knowledge point-level mapping relationship between the course content and the graph nodes through structural offset labels; S5: Generate a structural vector of the user input sentence based on syntactic parsing and semantic extraction, and compare it with the structural vector of the corresponding universal expression template in the graph to identify structural transfer errors. S6: Recommend teaching courses based on the context of the user's input sentence and the structural shift labels corresponding to structural transfer expression errors.

2. The method for recommending teaching courses based on English learning data according to claim 1, characterized in that: In S1, equivalent semantic expressions are extracted from the native language and English corpus, dependency syntactic analysis and structural component extraction are performed, and a set of cross-language structural difference features is generated, including: Perform semantic parsing on the aligned corpus of native language and English, and select sentence pairs with consistent meanings; Dependency structure analysis is performed on the native language expression and English expression in the sentence pairs with consistent sentence meaning, and a dependency structure tree is generated based on the inter-word dependency relationship; The syntactic backbone chain, intra-sentence modification relations and subordinate structure boundaries are extracted from the dependency structure tree, a set of syntactic structure components is established, and a syntactic function position alignment operation is performed on the set of structure components. The grammatical fragments with structural deviations in the expression are marked as structural difference fragments, and a set of language structure difference fragments is generated.

3. The method for recommending teaching courses based on English learning data according to claim 1, characterized in that: In S2, the set of structural difference segments is classified into multi-dimensional grammatical types, and a semantic-grammatical dimension mapping map is constructed, which specifically includes: Preset multi-dimensional grammatical structure offset tags, and divide the structural difference segments in the structural difference segment set into dimensions according to the structural offset tags to form offset segment groups; Semantic extraction is performed on the offset fragment group to form a joint semantic space, and an association mapping between the joint semantic space and the structural offset label is constructed. The structural offset label is used as the graph node and the joint semantic space is used as the connecting node to establish a semantic-syntactic dimension mapping graph.

4. The method for recommending teaching courses based on English learning data according to claim 1, characterized in that: In S3, a universal expression template is established for each connection node in the graph to generate a searchable structure vector. Specifically, the following steps are involved: In the semantic-syntactic dimension mapping graph, semantic item overlap analysis is performed on the joint semantic space corresponding to all connected nodes, and semantic items existing in multiple structural offset labels are integrated to construct a semantic intersection set. Using the semantic items in the semantic intersection set as index keys, aggregate all corresponding structural offset labels to generate a structural offset label group; According to the structure offset label group, the semantic items of the semantic intersection set are rewritten into a universal expression template based on the structure offset label and converted into a searchable structure vector.

5. The method for recommending teaching courses based on English learning data according to claim 4, characterized in that: Rewriting the semantic items of the semantic intersection set into a universal expression template based on the structure offset label and converting it into a searchable structure vector specifically includes: Obtain semantic items with the same structural offset tag group, obtain the syntactic backbone chain of the semantic items corresponding to the structural difference segments, and extract the corresponding word class arrangement rules of the backbone verbs, subjects, and core objects; According to the grammatical change mode specified by the current structure offset tag group, abstract placeholders are sequentially inserted into the generated syntactic backbone chain word order arrangement rules to form a general expression template grammar chain with an offset structure; The structural expression vector is constructed according to the three-dimensional factors of word class arrangement, structural offset label position, and label combination corresponding to the general expression template grammatical chain.

6. The method for recommending teaching courses based on English learning data according to claim 1, characterized in that: In S4, the teaching course is divided into knowledge point units according to the context scenario, and the course content is mapped to the graph nodes at the knowledge point level through the structural offset label. Specifically, Classify teaching courses according to actual communication context scenarios, label each course with the communication context label, and extract keywords for the knowledge point units of the teaching courses; Based on the grammatical structure offset label keywords corresponding to the graph nodes, the relationship mapping between the structure offset labels and the teaching course knowledge point units under different context labels is carried out through keyword matching, and a course recommendation decision engine based on the structure offset label is established according to the relationship mapping.

7. The method for recommending teaching courses based on English learning data according to claim 1, characterized in that: In S5, the structural vector of the user input sentence is generated based on syntactic parsing and semantic extraction, and is compared with the structural vector of the corresponding general expression template in the graph. The identification of structural transfer errors specifically includes: Perform word tagging and word dependency structure analysis on user input sentences based on grammatical structure offset labels, generate a dependency structure tree based on inter-word dependency relationships, and extract the syntactic backbone chain; Combine the syntactic backbone chain with the structural offset labeling of the word item to construct the structural vector of the input sentence; Perform semantic extraction on user input sentences, map the extraction results to the joint semantic space corresponding to the connected nodes in the semantic-syntactic dimension mapping graph, and perform semantic matching retrieval; If no matching semantic item is found, the user input statement is directly marked as grammatically incorrect; If a matching semantic item is successfully retrieved, the corresponding universal expression template structure vector is obtained and compared with the structure vector of the input sentence to identify structural transfer expression errors.

8. The method for recommending teaching courses based on English learning data according to claim 1, characterized in that: The vector comparison with the structural vector of the input sentence and the identification of structural transfer type expression errors are specifically performed by comparing the parameters in the vector item by item. If there are missing vector parameters or inconsistent orders, the input sentence is identified as a structural transfer type expression error.

9. The method for recommending teaching courses based on English learning data according to claim 1, characterized in that: In S6, based on the context of the user's input sentence and the structural shift labels corresponding to structural transfer errors, the following teaching courses are recommended: When a user input sentence is identified as a structural shift error, the context label corresponding to the semantic usage scenario of the input sentence is extracted and combined with the structural shift label combination in the general expression template as a course screening condition; The screening conditions are input into the course recommendation decision engine, and the final teaching course recommendation results are output.

10. A teaching course recommendation method system based on English learning data, used to implement the teaching course recommendation method based on English learning data according to any one of claims 1 to 9, characterized in that: It includes structure extraction module, graph construction module, template generation module, error recognition module and course recommendation module: Structure Extraction Module: This module performs dependency structure analysis on equivalent sentence pairs between the native language and English, extracts syntactic backbone chains, modification relations, and subordinate boundaries, and marks structural deviation segments according to component alignment to form language structure difference segments. Graph construction module: divides the structural difference segments into offset segment groups, extracts the semantic items within the groups to construct a joint semantic space, and establishes a semantic-syntactic dimension mapping graph with offset labels as nodes; Template generation module: Extracts the corresponding syntactic backbone chain sequence based on the structural offset tag group, embeds abstract placeholders to generate a universal expression template grammar chain, and generates a structural expression vector based on the word class sequence and tag combination; Error Identification Module: This module constructs structural vectors and extracts semantic terms from user sentences, searches the joint semantic space in the graph, and matches the common template vector. If parameters are missing or the order is shifted, it identifies a structural transfer error. Course recommendation module: collects input sentence context labels, combines them with error-corresponding structural offset label groups to form screening conditions, drives the course recommendation decision engine to match graph nodes with associated courses, and generates a target course recommendation list.

Images