A Data Query Method and Device Based on Terminology Anchoring

Abstract

This application discloses a data query method and apparatus based on terminology anchoring. The method includes: first, receiving user query data and the current session state; identifying the query data to obtain a final list of conceptual terms and a corresponding list of language decision tags, and obtaining a set of language tag knowledge blocks accordingly; combining the current session state, a preset bilingual terminology list, and the query data, reordering the knowledge block set using a scoring function; subsequently, updating the session-locked terms based on the reordering result to obtain a new session state; and finally, outputting the query answer based on the terminology list, language tags, reordered knowledge blocks, and the new session state. This method can enhance the accurate matching and standardized use of professional terms through terminology anchoring, avoiding problems such as terminology misuse and translation bias. It also achieves dynamic terminology locking and reuse by combining session state, improving the consistency, professionalism, and reliability of cross-language queries, and outputting more accurate and reliable results.

Description

Technical Field

[0001] This application relates to the field of data query technology, specifically to a data query method and apparatus based on term anchoring. Background Technology

[0002] In intelligent educational tutoring systems, cross-language retrieval is widely used to help second language learners acquire knowledge content in their native language or bilingual form. Existing methods typically employ multilingual embedding models for semantic matching, calculating the similarity between user queries and content blocks in the knowledge base to achieve cross-language knowledge retrieval and generation. However, in educational scenarios, this purely semantic matching approach often results in multiple translations of terms queried by users. Relying solely on semantic similarity makes it difficult for the system to distinguish the pedagogical applicability of these terms, maintain consistency in terminology usage within the same tutoring session, and may even retrieve knowledge blocks containing pseudo-cognates or literal translation errors. This can lead to inconsistent terminology, inappropriate translation choices, or misleading information in the system's output. Summary of the Invention

[0003] In view of the above problems, this application provides a data query method and apparatus based on term anchoring, which can solve the problems of existing methods relying solely on semantic similarity, resulting in confusing terminology in the system output, inappropriate translation selection, or the inclusion of misleading information.

[0004] Firstly, this application provides a data query method based on term anchoring, including: Receive query data and current session status input by the user; The query data is identified to obtain a final list of conceptual terms and a list of language decision tags corresponding to the final list of conceptual terms. Based on the final list of conceptual terms and the query data, obtain a set of language-marked knowledge blocks; Based on the current session state, the preset bilingual terminology list, and the query data, the language-marked knowledge block set is reordered using a scoring function to obtain a reordered knowledge block list; wherein, the standard term matching result and the alternative term matching result used in the scoring function are score-boosting items, and the cross-language trap detection result is a score-penalizing item; The current session state is updated with locked terms based on the reordered knowledge block list to obtain a new session state. The query answer is output based on the final list of conceptual terms, the list of language decision tags, the list of reordered knowledge blocks, and the locked terms in the new session state.

[0005] In the above technical solution, the method can enhance the accurate matching and standardized use of professional terms through the term anchoring mechanism, effectively avoiding the problems of term misuse, translation deviation and misleading information output caused by simply relying on semantic similarity. At the same time, it can realize the dynamic locking and continuous reuse of terms by combining the session state, improve the consistency, professionalism and reliability of cross-language data query, make the query results more in line with the expression norms of professional scenarios, and provide users with more accurate and reliable answers.

[0006] In some implementations, the step of identifying the query data to obtain a final list of conceptual terms and a list of language decision tags corresponding to the final list of conceptual terms includes: The query data is subjected to query concept term identification to obtain a final list of concept terms; The final conceptual terminology list is processed by language version tagging based on a preset bilingual terminology list to obtain a language decision tagging list; The language version is one of a first language version, a second language version, or a bilingual version.

[0007] In the above technical solution, the method can accurately identify the core concept terms in the query data and complete the standardized marking of the corresponding language versions according to the preset bilingual terminology list, providing an accurate basis for subsequent cross-language knowledge matching and terminology anchoring, and effectively ensuring the consistency and standardization of terminology recognition and language decision-making.

[0008] In some implementations, the step of performing language version tagging processing on the final concept term list according to a preset bilingual terminology list to obtain a language decision tag list includes: Obtain the user profile and conceptual terminology mastery level of the user; Based on the preset bilingual terminology list, obtain multiple difficulty ratings that correspond one-to-one with multiple concept terms in the final concept terminology list; The user's second language proficiency level is obtained based on the user profile, and the second language immersion coefficient is obtained based on the user profile. Based on the mastery level of the concept terms, obtain multiple current mastery levels corresponding one-to-one with multiple concept terms in the final list of concept terms; Based on the preset first weight parameter, the multiple difficulty ratings, the multiple current mastery levels, the second language proficiency level, and the second language immersion coefficient, calculate multiple confidence levels that correspond one-to-one with the multiple concept terms in the final list of concept terms; Based on the multiple confidence levels and preset judgment rules, language version tagging is performed on multiple concept terms in the final concept term list to obtain a language decision tag list; The determination rules include: Conceptual terms with a confidence level not less than a preset first confidence threshold are labeled as second languages; Conceptual terms with a confidence level not greater than a preset second confidence threshold are marked as first language conceptual terms; Conceptual terms with a confidence level less than the preset first confidence level threshold and greater than the preset second confidence level threshold are marked as bilingual conceptual terms; The first confidence threshold is greater than the second confidence threshold; The bilingual terminology maintains a terminology graph for each conceptual term, including nodes and edges; The nodes represent canonical terms, alternative terms, and trap patterns; The relationships represented by the edges include translation relationships, priority relationships, avoidance relationships, and context dependency relationships; The weights of the edges are used in the scoring function.

[0009] In the above technical solution, the method can dynamically calculate the confidence of terms by combining the user's language proficiency, terminology difficulty and mastery, realize the refined language version marking of conceptual terms, and rely on the structured terminology graph to comprehensively associate standard terms, alternative terms and trap patterns, providing accurate weight basis for subsequent scoring and re-ranking, and further improving the rationality and adaptability of cross-language terminology decision-making.

[0010] In some implementations, obtaining the language-marked knowledge block set based on the final list of conceptual terms and the query data includes: Obtain the set of concept term identifiers based on the final list of concept terms; A first knowledge block set is obtained based on a preset bilingual block index and the set of concept term identifiers; wherein, the list of concept terms corresponding to each knowledge block in the first knowledge block set includes at least one concept term identifier from the set of concept term identifiers; Based on the query data, a similarity search is performed in the preset bilingual block index to obtain the second knowledge block set; Merge the first knowledge block set and the second knowledge block set to obtain the initial knowledge block set; The initial set of knowledge blocks is deduplicated to obtain a set of associated knowledge blocks; The language version type of each knowledge block in the associated knowledge block set is marked according to the language decision type to obtain the language-marked knowledge block set.

[0011] In the above technical solution, the method can efficiently recall relevant knowledge blocks by combining precise matching and similarity retrieval. After deduplication, a concise and reliable candidate set is obtained, and the language version of the knowledge block is marked according to the language decision type, so as to provide high-quality and language-adapted knowledge materials for subsequent reordering and answer generation.

[0012] In some implementations, the step of reordering the language-tagged knowledge block set based on a scoring function according to the current session state, a preset bilingual terminology list, and the query data to obtain a reordered knowledge block list includes: Obtain the current lock term based on the session state; Calculate the semantic similarity between each knowledge block in the language-tagged knowledge block set and the query data; The terminology alignment score of each knowledge block in the language-marked knowledge block set is calculated based on a preset bilingual terminology list. Calculate the session consistency score for each knowledge block in the language-tagged knowledge block set based on the currently locked terminology; The score of each knowledge block in the language-marked knowledge block set is calculated based on the preset scoring function, the semantic similarity, the term alignment score, and the conversation consistency score. The language-marked knowledge block set is reordered based on the score to obtain a reordered knowledge block list. In the reordering process, when the score difference between multiple knowledge blocks is within a preset score threshold, a deterministic tie-breaking rule is used to determine the order. The order priority determined by the deterministic tie-breaking rule is as follows: alignment confidence, terminology standardization score, knowledge block length, and knowledge block identifier.

[0013] In the above technical solution, the method can comprehensively score semantic similarity, term alignment effect and conversation consistency in multiple dimensions, and ensure stable and reproducible ranking by breaking the rules through deterministic tie, so that the ranking of knowledge blocks is more in line with professional terminology norms and contextual coherence, and significantly improves the accuracy and orderliness of query results.

[0014] In some implementations, the language-marked knowledge block set is reordered based on the rating scores to obtain a reordered knowledge block list, including: The language-marked knowledge block set is reordered based on the scoring scores to obtain an initial sorted list; Based on a preset mastery control retrieval strategy, the initial sorted list is filtered by block type to obtain a reordered knowledge block list.

[0015] In the above technical solution, this method can further filter based on the score sorting and the user mastery strategy, so that the final returned knowledge blocks not only conform to the scoring sorting logic, but also adapt to the user's actual understanding ability, thereby improving the adaptability and usability of the query content while ensuring the accuracy and standardization of the results.

[0016] In some implementations, calculating the term alignment score of each knowledge block in the language-tagged knowledge block set based on a preset bilingual terminology list includes: Obtain a list of multiple concept terms that correspond one-to-one with multiple knowledge blocks in the language-marked knowledge block set; The preset bilingual terminology list is matched one by one with the multiple lists of conceptual terms to obtain matching results; wherein, the matching results include standard term matching results, alternative term matching results, cross-language trap detection results, and no matching results; Based on the matching results, determine multiple matching result assignment lists that correspond one-to-one with the multiple lists of concept terms; Calculate the term alignment score in the language tag knowledge block set that corresponds one-to-one with the multiple knowledge blocks according to the multiple matching result assignment lists; The assigned values ​​for the standard term matching result and the alternative term matching result are score boosting values, and the magnitude of the cross-language trap detection result is a score penalty value.

[0017] In the above technical solution, the method can complete the accurate matching of multiple types of conceptual terms in knowledge blocks by relying on bilingual terminology lists, distinguish different matching scenarios such as standard terms, alternative terms and cross-language traps, quantify the quality of term alignment through a reward and punishment assignment mechanism, accurately quantify the term compliance of each knowledge block, thereby effectively avoiding the erroneous information caused by cross-language term traps, and ensuring the professionalism, accuracy and rigor of knowledge block screening and scoring ranking.

[0018] In some embodiments, the method further includes: Cross-language trap patterns are obtained based on the bilingual terminology; wherein, the cross-language trap patterns are direction-sensitive, that is, the trap patterns from the first language to the second language and the trap patterns from the second language to the first language are directionally related; The query data is misunderstanding detected based on the cross-language trap pattern, and a list of misunderstandings is obtained and output. The misunderstanding list includes results where there is no misunderstanding or results where there is a misunderstanding; The existence of a misunderstanding includes at least one type of misunderstanding and a severity level corresponding to each type of misunderstanding; The types of misunderstandings include one or more of the following: pseudo-cognates, literal translation errors, terminology confusion, conceptual terminology transfer errors, and cross-linguistic symbolic convention differences.

[0019] In the above technical solution, the method can rely on the cross-language trap pattern with direction sensitivity to specifically detect various cross-language terminology misunderstandings in query data, accurately identify multiple types of misunderstandings such as false cognates, literal translation errors, and terminology confusion, and classify them into corresponding severity levels. This effectively identifies hidden errors and cognitive biases in cross-language queries in advance, avoids the generation of misleading information from the source, and thus improves the accuracy and reliability of cross-language data queries.

[0020] In some implementations, the step of updating the current session state with locked terms based on the reordered knowledge block list to obtain a new session state includes: Obtain the first knowledge block that is ranked first from the reordered knowledge block list, and obtain the concept term associated with the first knowledge block; When the current session state does not include the concept term associated with the first knowledge block, the concept term associated with the first knowledge block is identified as a new locked term; The current session state is updated according to the new locking term to obtain the new session state.

[0021] In the above technical solution, the method can dynamically filter high-quality core concept terms based on the knowledge block sorting results, complete the iterative update of session-locked terms, realize the continuous optimization and solidification of the state of session context terms, ensure the uniformity and continuity of term usage within the same session, avoid term conflicts and expression confusion in the context, and continuously improve the standardization and consistency of cross-language query responses.

[0022] Secondly, this application provides a data query device based on term anchoring, comprising: The receiving unit is used to receive query data input by the user and the current session status; The identification unit is used to identify the query data to obtain a final list of concept terms and a list of language decision tags corresponding to the final list of concept terms; The acquisition unit is used to acquire a set of language mark knowledge blocks based on the final list of conceptual terms and the query data; The reordering unit is used to reorder the language-marked knowledge block set based on a scoring function according to the current session state, a preset bilingual terminology list, and the query data to obtain a reordered knowledge block list; wherein, the standard term matching result and the alternative term matching result used in the scoring function are score-boosting items, and the cross-language trap detection result is a score-penalizing item. The update unit is used to update the current session state with locked terms based on the reordered knowledge block list to obtain a new session state; The output unit is used to output the query answer based on the final conceptual terminology list, the language decision tag list, the reordered knowledge block list, and the locked terms in the new session state.

[0023] In the above technical solution, the device can enhance the accurate matching and standardized use of professional terms through the term anchoring mechanism, effectively avoiding the problems of term misuse, translation deviation and misleading information output caused by simply relying on semantic similarity. At the same time, it can realize the dynamic locking and continuous reuse of terms by combining the session state, improve the consistency, professionalism and reliability of cross-language data query, make the query results more in line with the expression norms of professional scenarios, and provide users with more accurate and credible answers.

[0024] Thirdly, this application provides an electronic device including a memory and a processor, the memory storing a computer program, and the processor running the computer program to cause the electronic device to perform the term-anchored data query method described in any one of the first aspects.

[0025] Fourthly, this application provides a readable storage medium storing a computer program, which, when executed by a processor, performs the term-anchored data query method described in any one of the first aspects.

[0026] Fifthly, this application provides a computer program product comprising a computer program that, when executed by a processor, performs the term-anchored data query method described in any one of the first aspects.

[0027] The beneficial effects of this application are: it can effectively improve the consistency of terminology in cross-language tutoring, reduce learners' confusion caused by translation inconsistencies, achieve adaptive language adaptation that fits specific concepts and learners' levels, and identify and detect various cognitive misunderstandings caused by translation in advance. Attached Figure Description

[0028] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0029] Figure 1This is a flowchart illustrating a data query method based on term anchoring in some embodiments of this application; Figure 2 This is a schematic diagram of the structure of a term-anchored data query device in some embodiments of this application; Figure 3 This is a schematic diagram of the structure of an electronic device in some embodiments of this application. Detailed Implementation

[0030] The embodiments of the technical solution of this application will now be described in detail with reference to the accompanying drawings. These embodiments are only used to more clearly illustrate the technical solution of this application and are therefore merely examples, and should not be used to limit the scope of protection of this application.

[0031] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the application; the terms “comprising” and “having”, and any variations thereof, in the specification, claims, and foregoing description of the drawings are intended to cover non-exclusive inclusion.

[0032] In the description of the embodiments of this application, technical terms such as "first" and "second" are used only to distinguish different objects and should not be construed as indicating or implying relative importance or implicitly specifying the number, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this application, "multiple" means two or more (including two), similarly, "multiple sets" refers to two or more sets (including two sets), and "multiple pieces" refers to two or more pieces (including two pieces) unless otherwise explicitly defined.

[0033] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0034] In the description of the embodiments in this application, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0035] This application addresses the problems of inconsistent terminology translation, lack of pedagogical constraints in semantic retrieval, inability to adapt to learners' language proficiency, and susceptibility to cross-language misunderstandings in existing cross-language retrieval enhancement generation systems in educational scenarios. Based on the concept of terminology anchoring in cognitive science, this application proposes a cross-language retrieval enhancement generation scheme for educational scenarios.

[0036] This method uses teaching concepts as the core for terminology anchoring retrieval, relies on bilingual terminology lists to constrain and reorder retrieval results, adaptively selects the presentation language based on concept difficulty and learner level, maintains terminology consistency during dialogue, and actively detects and handles cross-language misunderstandings, thereby achieving more accurate, standardized and teaching-adaptive cross-language knowledge retrieval and tutoring.

[0037] like Figure 1 As shown, some embodiments of this application provide a term-anchored data query method, which includes: S100: Receive query data and current session status input by the user.

[0038] In this embodiment, the system acquires real-time query text data input by the user and conversation state information stored in historical dialogues, providing basic input data for subsequent full-process processing.

[0039] In this embodiment, query data refers to the natural language text content that the user currently inputs into the system to initiate a knowledge query. That is, the original text of the question raised by the user or the knowledge point that needs to be explained. It is the original input object for the system to perform concept recognition, terminology retrieval and answer generation.

[0040] For example, the query data can be teaching-related questions actually entered by users, such as "What is a derivative?", "How to calculate the limit of a function?", and "Please explain the difference between differential and derivative".

[0041] In this embodiment, the current session state refers to the context information maintained by the system during this dialogue, which mainly includes the concepts identified in this round of the conversation, the defined standard terms, the locked terminology set, and historical language decision records, in order to ensure the consistency of terminology and the stability of expression within the conversation.

[0042] For example, the current session state can be contextual information such as: the system has locked the corresponding specification term for "derivative" as "derivative", the identified concepts include "limit" and "function", and the current explanation language is mainly Chinese.

[0043] S200. Identify the query data to obtain the final list of concept terms and the corresponding list of language decision tags.

[0044] In this embodiment, the final list of concept terms refers to the set of standardized concept terms that are extracted from user query data through standardized term matching, alternative term matching and fuzzy matching, and correspond one-to-one with teaching knowledge points. Each concept has a unique identifier and professional meaning.

[0045] In this embodiment, the language decision label list refers to the set of language type labels obtained by adaptively determining each concept in the final concept term list based on concept difficulty, learner's second language level, concept mastery and learning objectives. The language type includes first language, second language or bilingual.

[0046] In this embodiment, the first language can be Chinese, and the second language can be English. The language decision tag list can be understood as indicating the language to be used for the final answer output. S300. Based on the final list of conceptual terms and query data, obtain the set of language-tagged knowledge blocks.

[0047] In this embodiment, the system can first perform precise retrieval based on conceptual terms, and then perform semantic similarity retrieval in combination with query data. The two types of retrieval results are merged and deduplicated to obtain a set of candidate knowledge blocks related to the concept.

[0048] In this embodiment, the language-tagged knowledge block set refers to the set of candidate knowledge blocks that are recalled from the bilingual block index based on concept identifiers and semantic similarity, and whose language version is marked according to the language decision results. Each knowledge block carries information on its associated concept, language type, terminology standardization, and alignment confidence.

[0049] In this embodiment, the language knowledge block set can be understood as a standardized teaching knowledge content that is related to the query concept and has been annotated in the appropriate language.

[0050] S400. Based on the current session state, the preset bilingual terminology list, and the query data, the language-marked knowledge block set is reordered according to a scoring function to obtain a reordered knowledge block list. The standard term matching results and alternative term matching results used in the scoring function are score-boosting items, and the cross-language trap detection results are score-penalizing items.

[0051] In this embodiment, the pre-built bilingual terminology is a standardized professional lexicon and concept relation database that is built and stored in advance. It uniformly includes concept identifiers, domain classifications, difficulty levels, bilingual standard terms, alternative terms, cross-language traps, related concepts and prerequisite concepts of various disciplines. At the same time, it constructs a terminology graph to represent the relationship between terms, providing a unified standard basis for terminology comparison, scoring constraints and language decision-making.

[0052] For example, each entry in the bilingual terminology list contains: concept_id: A unique identifier for a concept (e.g., 'CALC_DERIVATIVE_001'); domain: field (e.g., 'calculus'). difficulty: difficulty rating ∈ [0,1]; en / zh: Standard terminology, alternative terms, definitions, and common misunderstandings for each language; cross_lingual_traps: Cross-lingual trap patterns (false cognates, translation errors); related_concepts: Links to related concepts; prerequisite_concepts: prerequisite concepts.

[0053] For example, in the fields of calculus and linear algebra, the bilingual glossary contains approximately 500 core concepts. Each concept has one standard term and 2–5 alternative terms in each language. In addition, the bilingual glossary also identifies approximately 150 easily confused pseudo-cognates and typical translation errors, and summarizes approximately 300 common misunderstandings caused by language differences.

[0054] In this embodiment, the reordered knowledge block list is an ordered list of content formed after scoring and prioritizing the original set of language-marked knowledge blocks. It is arranged from high to low according to the comprehensive score, providing preferred knowledge materials for subsequent answer output.

[0055] In this embodiment, the scoring function is a quantitative calculation model that integrates multi-dimensional indicators. It can combine semantic similarity, term alignment score, and conversation consistency score for weighted calculation to objectively quantify the overall quality of each knowledge block.

[0056] In this embodiment, the standardized terminology matching result refers to the complete matching of the knowledge block content with the official standard and uniformly prescribed professional terms in the bilingual terminology table, which is the core basis for determining the professionalism of the content; The alternative term matching result refers to the matching situation formed by non-standard synonyms that are compliant, general, and equivalently replaceable under the same concept in knowledge blocks, which belongs to the reasonable and compliant form of term use; The cross-language trap detection results are based on the directed trap pattern recognition of the glossary, identifying cross-language comprehension misunderstandings and erroneous expressions such as false cognates, literal translation errors, and terminological confusion.

[0057] In this embodiment, the score enhancement item refers to a positive matching indicator that can prove that the content of the knowledge block is standardized, professional, and compliant. After being included in the calculation, it can increase the comprehensive score of the knowledge block and improve the content ranking priority. The penalty items refer to negative indicators that indicate problems such as misunderstanding, translation errors, or misuse of terminology. When included in the scoring calculation, the score of the knowledge block will be deducted to reduce the display priority of the problem content.

[0058] For example, the scoring function can use semantic similarity as the basic benchmark score, combine positive indicators of canonical term matching and alternative term matching for score optimization, and deduct points based on the error results of cross-language trap detection. At the same time, it can combine the conversation consistency score for comprehensive weighting to finally complete the quantitative scoring of the quality of knowledge blocks.

[0059] S500: Update the locked terms of the current session state according to the reordered knowledge block list to obtain the new session state.

[0060] In this embodiment, the method follows the session terminology locking rule: prioritize locking the normative concept terms associated with high-priority knowledge blocks, solidify the terminology usage standard within the same session, and only support active unlocking and adjustment.

[0061] Implementing this method can effectively avoid problems such as mixed terminology and inconsistent expressions in the conversation.

[0062] S600: Output the query answer based on the final concept terminology list, language decision tag list, reordered knowledge block list, and locked terms in the new session state.

[0063] In this embodiment, the method can integrate concept recognition results, language selection strategies, high-quality knowledge content and conversation-locked terms, fuse cross-language misunderstanding detection results, and output standardized terminology answers, misunderstanding prompts and standardized explanations in one integrated manner.

[0064] For example, if the final list of conceptual terms is derivative and limit, the language decision tags are bilingual display of derivative and Chinese display of limit, and the conversation has locked the standard term "derivative", the system will output the standard answer in the corresponding language paradigm based on the reordered high-quality knowledge blocks.

[0065] In the above embodiments, the method can enhance the accurate matching and standardized use of professional terms through the term anchoring mechanism, effectively avoiding the problems of term misuse, translation deviation and misleading information output caused by simply relying on semantic similarity. At the same time, it can realize the dynamic locking and continuous reuse of terms by combining the session state, improve the consistency, professionalism and reliability of cross-language data query, make the query results more in line with the expression norms of professional scenarios, and provide users with more accurate and credible answers.

[0066] In some embodiments, after step S100, the method may further include: S700. Obtain cross-language trap patterns based on the bilingual glossary; wherein, the cross-language trap patterns are directionally sensitive, that is, the trap patterns from the first language to the second language and the trap patterns from the second language to the first language are directionally related.

[0067] In this embodiment, the bilingual terminology includes a massive set of directional cross-language trap records, covering various error patterns such as pseudo-cognates, literal translation deviations, and symbol differences. These are used to distinguish the differentiated trap features in bidirectional translation scenarios and provide rule support for accurate misunderstanding detection.

[0068] S800: Perform misunderstanding detection on the query data based on the cross-language trap pattern, obtain the misunderstanding list and output it.

[0069] In this embodiment, the misunderstanding list includes results that have no misunderstanding or results that have a misunderstanding; The existence of a misunderstanding includes at least one type of misunderstanding and the severity level corresponding to each type of misunderstanding; Misunderstanding types include one or more of the following: pseudo-cognates, literal translation errors, terminology confusion, conceptual term transfer errors, and cross-linguistic symbolic convention differences.

[0070] In this embodiment, cross-language misunderstanding detection covers five typical problems: (1) False cognates: words that are similar in appearance or pronunciation but have different meanings, such as mistranslating “derivative” in mathematics as “derivative” in chemical terms; (2) Literal translation error: refers to the loss of professional semantics due to word-for-word translation, such as mistranslating the mathematical term "function" as the general meaning of "function"; (3) Confusing terms: refers to the use of multiple equivalent terms in the expression of the same concept, such as alternating between "derivative" and "differential coefficient" in the same content; (4) Concept transfer error: refers to directly applying the conceptual structure of the mother tongue to the second language, such as directly using Chinese word order in English expression; (5) Symbol mismatch: refers to different symbol systems used in different language textbooks, such as different variable naming conventions used in Chinese and English textbooks.

[0071] In this embodiment, the misunderstanding detection algorithm used in this method can identify false cognates and language traps by matching bilingual terminologies, while also detecting internal inconsistencies caused by the mixing of multiple terms for the same concept, and mapping severity levels based on scores: A score of <0.5 indicates a minor severity level (MINOR). A score of 0.5 or less and less than 0.8 is considered moderate (corresponding to a severity level of MODERATE). A score of ≥0.8 indicates a severe condition (corresponding to a severity level of Severe).

[0072] Based on this, the method can automatically identify various cross-language comprehension deviations by relying on the built-in misunderstanding database and grading standards in the glossary, assess the degree of harm of the misunderstandings according to three levels: minor, moderate and severe, and can be linked with the learning tracking module to complete the recording, tracing and precise intervention of misunderstandings.

[0073] For example, pseudo-cognates and literal translation errors are marked as moderate severity, and a blocking / unblocking strategy is implemented; Terminology confusion is marked as mild severity, and only a warning is given without any blocking action.

[0074] In this embodiment, the misunderstanding list obtained by misunderstanding detection can be connected to an external misunderstanding tracker, so that the misunderstanding tracker can record the status of the misunderstanding information recorded in the misunderstanding list, maintain its life cycle, and control the unlocking linkage, thereby realizing the full life cycle management of misunderstandings.

[0075] Specifically, the misunderstanding tracker can track the lifecycle status of each misunderstanding in the misunderstanding list, including statuses such as detected, active, processed, and cleared. Among them, a blocking and unlocking condition is set for misunderstandings in the active state. The blocking and unlocking condition is that the misunderstanding will be cleared and removed from the misunderstanding list only when there are two consecutive rounds of no misunderstanding mode and the micro-quiz corresponding to the relevant concept and term is correct.

[0076] For example, for cases marked as moderately severe false cognates or literal translation errors in the misunderstanding list, the misunderstanding tracker implements a blocking unlocking control strategy, that is, it prohibits the unlocking output of the complete solution content; For cases of terminology confusion marked as minor severity in the misunderstanding list, the misunderstanding tracker only implements a warning control strategy, that is, it only outputs a reminder message without performing any content unlocking or blocking processing.

[0077] In the above embodiments, the method can rely on a cross-language trap pattern with direction sensitivity to specifically detect various cross-language terminology misunderstandings in query data, accurately identify multiple types of misunderstandings such as false cognates, literal translation errors, and terminology confusion, and classify them into corresponding severity levels. This effectively identifies hidden errors and cognitive biases in cross-language queries in advance, avoids the generation of misleading information from the source, and thus improves the accuracy and reliability of cross-language data queries.

[0078] In some embodiments, step S200 may include: S210. Identify query concepts and terms in the query data to obtain the final list of concepts and terms.

[0079] In this embodiment, the method adopts a three-stage progressive concept recognition scheme, which sequentially completes concept extraction through precise matching of standardized terms, weighted matching of alternative terms, and embedded fuzzy matching. A fixed fuzzy matching threshold is set to balance recognition accuracy and recall.

[0080] For example, the first stage is exact matching of canonical terms with a confidence level of 1.0; The second stage is alternative term matching, where the confidence level is determined by the term weight, typically 0.7. 0.9; The third stage is embedding-based fuzzy matching. As a fallback strategy, the fuzzy matching threshold θ_fuzzy∈[0.75,0.90] has a default value of 0.80.

[0081] The confidence score is used to characterize the credibility of the concept identification results. The closer the score is to 1, the more standard and reliable the identification result is. Based on this, the confidence score is fixed at 1.0 in the first stage, indicating that the query text completely matches the standard terminology in the glossary, and the identification result is absolutely credible and unbiased. The second stage uses a confidence score in the range of 0.7 to 0.9, indicating that the query matches compliant alternative terms, which are conceptually consistent but not official standard terms, so the confidence weight is appropriately reduced. The third stage sets a fuzzy matching threshold of 0.75 to 0.90, with a default of 0.80, meaning that a match is only considered successful when the semantic similarity reaches the threshold standard. This is used as a fallback to identify non-standard, colloquial, and variant expressions in the query content, minimizing missed concept detection while ensuring recognition accuracy.

[0082] S220. Based on the preset bilingual terminology list, the final concept terminology list is processed by language version marking to obtain a language decision mark list.

[0083] In this embodiment, the language version is one of the following: a first language version, a second language version, or a bilingual version.

[0084] For example, the first language version can be Chinese, and the second language version can be English.

[0085] In this embodiment, the method can pre-organize and collect professional vocabulary in teaching scenarios, cross-language comparison relationships, easily confused and trap words, and synonym substitution terms; pre-sort out the single language usage norms, bilingual adaptation rules and language selection constraints corresponding to each concept term, and complete the pre-definition and binding of term language attributes, applicable scenarios and language limitation rules, thereby completing the pre-construction of the bilingual terminology list and providing a standardized basis for the subsequent automated language version marking of batch terms.

[0086] For example, a bilingual glossary of calculus concepts can be shown in Table 1.

[0087] Table 1. Bilingual Glossary of the Concept of Calculus

[0088] For example, a bilingual glossary of linear algebra concepts can be shown in Table 2.

[0089] Table 2. Bilingual Glossary of Linear Algebra Concepts

[0090] To improve the rationality and adaptability of cross-linguistic terminology decision-making, step S220 may further include: S221. Obtain user profiles and understanding of conceptual terminology.

[0091] In this embodiment, the method connects with learning data to obtain learners' personal ability profiles and long-term mastery data of individual professional concepts in real time, providing a basis for personalized language selection.

[0092] In this embodiment, the optimal language choice for bilingual learners depends on: (1) Conceptual difficulty: Concepts with high difficulty may benefit from L1 (first language) explanation; (2) Learner's L2 (second language) level: Higher level refers to more L2-beneficial experiential learning; (3) Concept-specific level: Students may understand some concepts better in L1 and other concepts better in L2 (English); (4) Learning objectives: Immersive learning or comprehension-first learning; In this example, L1 represents the first language and L2 represents the second language. In a Chinese-English bilingual scenario, L1 can be Chinese and L2 can be English; in a Chinese-Japanese bilingual scenario, L1 can be Chinese and L2 can be Japanese. This embodiment does not impose any limitations on this. This embodiment will be described using a Chinese-English bilingual scenario as an example.

[0093] In this embodiment, immersive learning refers to a learning model centered on second language exposure to professional knowledge, prioritizing English for content explanation and knowledge input, and strengthening second language application ability through a fully foreign language context; comprehension-first learning refers to a learning model with the primary goal of thoroughly understanding and mastering knowledge points, prioritizing clear explanations of complex and difficult concepts in Chinese, lowering the language threshold, and ensuring accurate understanding of knowledge.

[0094] S222. Obtain multiple difficulty ratings corresponding to multiple concept terms in the final concept term list based on the preset bilingual term list.

[0095] In this embodiment, the bilingual glossary can maintain a terminology graph for each conceptual term, including nodes and edges; Nodes represent canonical terminology, alternative terminology, and trap patterns; The relationships represented by edges include translation relations, priority relations, avoidance relations, and context dependency relations; The edge weights are used in the scoring function.

[0096] In this embodiment, the method can maintain a terminology graph for each concept based on a bilingual terminology table (such as a flat JSON entry table).

[0097] For example, nodes represent canonical terms, alternative terms, and trap patterns. The relationships represented by edges include: TRANSLATION, PREFER_OVER, AVOID, and CONTEXT_DEPENDENT. Among them, translation relations represent cross-language equivalence and are used to improve the score of knowledge blocks with aligned translations; Priority relations indicate contextual preferences and are used to improve the score of preferred terms in the matching context; Avoidance relationships correspond to known traps or error patterns and are used to penalize knowledge blocks according to severity weights; Context-dependent relationships mean that multiple expressions are valid, depending on the context, and maintain a neutral score when there is no clear context signal.

[0098] In this embodiment, the terminology list is configured with a quantitative difficulty rating of 0 to 1 for each concept. Combined with the weight parameters of different relation edges in the terminology graph, it provides quantitative support for language decision-making, terminology scoring and content selection.

[0099] In this embodiment, the terminology graph is used to structurally represent the relationships between terms, such as translation, priority, avoidance, and contextual dependence, providing underlying support for terminology recognition and language decision-making.

[0100] S223. Obtain the user's second language proficiency level based on the user profile, and obtain the second language immersion coefficient based on the user profile.

[0101] In this embodiment, learners' second language proficiency and learning goal preferences are extracted from user profiles to distinguish the needs of different learning scenarios such as immersive learning and comprehension-first learning.

[0102] In this embodiment, the higher the level of second language proficiency, the more suitable it is to use more second language for experiential learning; the learning objective is used to identify whether to emphasize second language immersion.

[0103] S224. Based on the mastery of concept terms, obtain multiple current mastery levels that correspond one-to-one with multiple concept terms in the final list of concept terms.

[0104] In this embodiment, based on long-term learning behavior data, the mastery score of each concept is quantitatively output, which accurately reflects the learner's weak understanding of different professional knowledge.

[0105] In this embodiment, the concept mastery level is derived from the AI ​​education tutoring response control system and method based on the knowledge mastery level gating mechanism, and is used to characterize the learner's understanding of the current concept.

[0106] S225. Based on the preset first weight parameter, multiple difficulty ratings, multiple current mastery levels, second language proficiency level, and second language immersion coefficient, calculate multiple confidence levels that correspond one-to-one with multiple concept terms in the final list of concept terms.

[0107] In this embodiment, a language selection calculation model is constructed based on the sigmoid activation function, which integrates multi-dimensional features and fixed weight parameters to quantify the probability value of each concept being displayed in a second language.

[0108] In this embodiment, the formula for calculating the language selection probability that corresponds one-to-one with multiple concept terms in the final list of concept terms is as follows: ; Where σ is the sigmoid activation function, used to normalize and map the multidimensional weighted calculation results to the standard probability interval of 0 to 1; difficulty∈[0,1] is the concept difficulty coefficient taken from the bilingual terminology list. The larger the value, the higher the difficulty of understanding the current professional concept. proficiency_L2∈[0,1] is a quantitative value of the learner's second language proficiency, used to characterize the learner's ability to receive and understand second language knowledge content; mastery_k∈[0,1] represents the learner's current mastery of the k-th concept. This data is obtained by comprehensively calculating the learner's historical interaction evidence, knowledge forgetting decay, learning performance, and misunderstanding status, and is used to quantitatively reflect the learner's solid knowledge of the corresponding concept. goal_L2∈{0,1} is the learning objective labeling parameter. A value of 1 indicates that the learning objective is second language immersion learning, and a value of 0 indicates that the learning objective is comprehension-oriented learning. w_d is a preset weight coefficient corresponding to the concept difficulty, which is used to adjust the weight of the influence of the concept difficulty factor on the language selection probability. w_p is a preset weighting coefficient corresponding to the learner's second language proficiency, which is used to adjust the weight of the learner's second language ability on the probability of language selection. w_m is a preset weighting coefficient corresponding to the concept mastery level, which is used to adjust the weight of the influence of the student's current knowledge level on the probability of language selection. w_g is the preset weight coefficient corresponding to the learning objective, which is used to adjust the weight of the influence of the immersive learning objective on the language selection probability.

[0109] In this embodiment, the determination rules include: Concepts and terms with a confidence level not less than a preset first confidence level threshold (0.6) are marked as second languages; Conceptual terms with a confidence level not greater than the preset second confidence level threshold (0.4) are marked as first language conceptual terms; Conceptual terms with a confidence level less than a preset first confidence level threshold and greater than a preset second confidence level threshold are marked as bilingual conceptual terms; The first confidence level is greater than the second confidence threshold.

[0110] For example, the decision logic is as follows: If P(L2)≥0.6: Return 'L2' (in English); If P(L2)≤0.4: return 'L1' (in Chinese); Otherwise: Return to 'BILINGUAL' (bilingual presentation).

[0111] In this embodiment, the dual threshold judgment logic is strictly followed, and the concept difficulty, students' level and mastery are combined to adaptively match monolingual or bilingual display schemes for different concepts, so as to achieve personalized teaching language adaptation.

[0112] For example, language selection decisions can be made according to the following example rules: (1) When the concept difficulty is high (0.8), the L2 level is low (0.3), and the mastery is low (0.3), the decision is to use L1 (Chinese). The principle is that the concept difficulty is high and the learner's second language ability is weak, so the native language should be used for explanation first.

[0113] (2) When the concept difficulty is high (0.8), the L2 level is high (0.8), and the mastery is low (0.3), the decision is to use bilingualism. The principle is that the concept difficulty is high, but the learner has a certain second language foundation, and the understanding can be enhanced through two languages.

[0114] (3) When the concept difficulty is low (0.2), the L2 level is low (0.3), and the mastery is high (0.7), the decision is to use L2 (English). The principle is that the concept is simple and the learner has mastered the knowledge points, so the second language learning can be promoted appropriately.

[0115] (4) When the concept difficulty is medium (0.5), the L2 level is medium (0.5), and the mastery is medium (0.5), the decision is to adopt bilingualism. The principle is that the indicators are moderate and bilingual presentation is used to balance understanding and language learning.

[0116] (5) When the concept difficulty is low (0.2), the L2 level is high (0.8), and the mastery is arbitrary, the decision is made to use L2 (English). The principle is that the concept is simple and the second language ability is strong, which is suitable for immersive learning.

[0117] For example, when the concept is difficult, the second language proficiency is low, and the mastery is low, the first language is used; when the difficulty is high, the second language proficiency is high, but the mastery is low, bilingualism is used; when the difficulty is low and the second language proficiency is high, the second language is used directly; when the difficulty, proficiency, and mastery are all moderate, a balanced bilingual presentation is used.

[0118] In the above embodiments, the method can accurately identify the core concept terms in the query data and complete the standardized marking of the corresponding language versions according to the preset bilingual terminology list, providing an accurate basis for subsequent cross-language knowledge matching and terminology anchoring, and effectively ensuring the consistency and standardization of terminology recognition and language decision-making.

[0119] In some embodiments, step S300 may include: S310. Obtain the set of concept term identifiers based on the final list of concept terms.

[0120] In this embodiment, a standardized set of concept_ids is extracted based on the unique concept numbers compiled in a unified glossary, so as to achieve precise binding and association between knowledge blocks and professional concepts.

[0121] In this embodiment, the concept term identifier set refers to a collection of unique identifiers assigned to each professional concept term, using concept_id as the basic unit. Different knowledge points correspond to independent and non-repeating identifiers. This set is used to convert textual concept terms into machine-readable encoded forms, facilitating subsequent rapid retrieval, accurate matching, and categorized access to corresponding teaching knowledge content.

[0122] S320. Obtain a first knowledge block set based on a preset bilingual block index and a set of concept term identifiers; wherein, the list of concept terms corresponding to each knowledge block in the first knowledge block set includes at least one concept term identifier from the set of concept term identifiers.

[0123] In this embodiment, the method utilizes the concept association retrieval capability of bilingual block indexes to filter professional knowledge blocks that are strongly bound to the target concept, thereby ensuring the professional relevance of the retrieved content.

[0124] In this embodiment, each knowledge block in the preset bilingual block index has a standard block structure, including: Chunk unique identifier: chunk_id; A list of associated concept IDs: concept_ids; Content type: chunk_type; English and Chinese text content: content.en / content.zh; Alignment level between English and Chinese versions: alignment_confidence; Terminology standardization level: term_canonical_score.

[0125] In this embodiment, the value range of alignment_confidence is [0.7, 0.95], which is used to represent the alignment confidence of Chinese and English content; The value of term_canonical_score ranges from [0.7, 1.0] and is used to indicate the degree of standardization of terms within a knowledge block.

[0126] In this embodiment, the method pre-divides all teaching-related content into standardized knowledge blocks, and uniformly labels them according to the content purpose and display format of the knowledge blocks to form corresponding content types; The content type (chunk_type) includes: DEFINITION, EXAMPLE, HINT, PARTIAL, WORKED, and ANSWER.

[0127] Based on this, different content types can be used to distinguish the display format and open strategy of different knowledge blocks, thereby providing a classification basis for subsequent knowledge block filtering and content control based on mastery status and misunderstanding detection results.

[0128] In this embodiment, the system uses a set of preset parameters for retrieval and sorting, including: Semantic weight α, term alignment weight β, consistency weight γ, number of search candidates top_k, final number of returned blocks N, fuzzy matching threshold θ_fuzzy, high second language probability threshold θ_L2_hi, and low second language probability threshold θ_L2_lo.

[0129] In this embodiment, the semantic weight α ranges from [0.3, 0.7], with a default value of 0.5, and is used to represent the degree of contribution of embedding similarity; The term alignment weight β ranges from [0.2, 0.5], with a default value of 0.3, and is used to represent the anchoring strength of the term list; The consistency weight γ ranges from [0.1, 0.3], with a default value of 0.2, and is used to represent the importance of session coherence.

[0130] In this embodiment, the value range of the search top_k is [20, 100], and the default value is 50, which represents the number of candidate blocks in the initial semantic search; The final returned number N has a range of [3, 10], with a default value of 5, which represents the number of knowledge blocks returned after reordering.

[0131] In this embodiment, the fuzzy matching threshold θ_fuzzy ranges from [0.75, 0.90], with a default value of 0.80, and is used to control the activation condition of embedded backoff matching; The high threshold value θ_L2_hi for second language probabilities ranges from [0.55, 0.70], with a default value of 0.60; if P(L2) ≥ this value, select L2. The second language probability low threshold θ_L2_lo has a value range of [0.30, 0.45], with a default value of 0.40; if P(L2) ≤ this value, select L1.

[0132] S330. Based on the query data, perform a similarity search in the preset bilingual block index to obtain the second knowledge block set.

[0133] In this embodiment, the whole-domain semantic similarity retrieval is completed by relying on the embedded vector, which makes up for the coverage shortcomings of precise concept retrieval and improves the comprehensiveness of candidate knowledge.

[0134] S340. Merge the first knowledge block set and the second knowledge block set to obtain the initial knowledge block set.

[0135] In this embodiment, the results of precise retrieval and semantic retrieval are integrated to expand the candidate knowledge pool and balance retrieval accuracy and content richness.

[0136] S350. Perform deduplication on the initial knowledge block set to obtain the associated knowledge block set.

[0137] In this embodiment, the method eliminates duplicate and redundant knowledge content, simplifies the candidate set, and reduces the computational overhead of subsequent reordering.

[0138] S360. Label the language version type of each knowledge block in the associated knowledge block set according to the language decision type to obtain the language-labeled knowledge block set.

[0139] In this embodiment, the candidate knowledge blocks are screened and labeled with language based on the results of the previous conceptual language decision-making, and matched with the learner's personalized language reception habits.

[0140] As can be seen, the above process describes a method that first retrieves data using concept_id (high precision); then retrieves data using semantic similarity (high recall); and finally deduplicates the data, retaining the set of language-tagged knowledge blocks with the highest confidence.

[0141] In the above embodiments, the method can efficiently recall relevant knowledge blocks by combining precise matching and similarity retrieval, obtain a concise and reliable candidate set after deduplication, and complete the language version labeling of knowledge blocks according to language decision type, providing high-quality, language-adapted knowledge materials for subsequent reordering and answer generation.

[0142] In some embodiments, step S400 may include: S410. Obtain the current lock term based on the session state.

[0143] In this embodiment, locked terminology data stored in historical sessions is read and parsed as a benchmark for session consistency scoring.

[0144] S420, Computational language tagging of the semantic similarity between each knowledge block in the knowledge block set and the query data.

[0145] In this embodiment, the embedding model is called to calculate the semantic matching degree between the question and the knowledge block text, and a standardized score in the range of 0 to 1 is output as a basic relevance evaluation index.

[0146] S430. Calculate the term alignment score for each knowledge block in the language-tagged knowledge block set based on the preset bilingual terminology list.

[0147] In this embodiment, the compliance and accuracy of terminology usage within a knowledge block are quantitatively evaluated based on the glossary of standardized entries and constraint rules.

[0148] To ensure the professionalism, accuracy, and rigor of the knowledge block selection and scoring ranking, step S430 may also include: S431. Obtain a list of multiple concept terms that correspond one-to-one with multiple knowledge blocks in the language tag knowledge block set.

[0149] In this embodiment, all the professional concepts and terms contained in each knowledge block are broken down to establish basic data for comparison.

[0150] S432. Match the preset bilingual terminology list with multiple conceptual terminology lists one by one to obtain matching results; wherein, the matching results include standard terminology matching results, alternative terminology matching results, cross-language trap detection results, and no matching results.

[0151] In this embodiment, the assignment results corresponding to the standard term matching result and the alternative term matching result are score boosting values, and the amplitude result corresponding to the cross-language trap detection result is a score penalty value.

[0152] In this embodiment, different scoring rules are set for different term matching types. Standardized terms and compliant alternative terms are given positive scores, while cross-language traps and incorrect expressions are penalized according to their severity.

[0153] S433. Determine multiple matching result assignment lists that correspond one-to-one with multiple lists of concept terms based on the matching results.

[0154] In this embodiment, the scores of various matching results are uniformly quantified to form a standardized value list, which facilitates unified calculation.

[0155] S434. Calculate the term alignment score in the language tag knowledge block set that corresponds one-to-one with multiple knowledge blocks according to the multiple matching result assignment lists.

[0156] In this embodiment, the score normalization process is completed by combining the number of terms, and the term alignment score in the range of -1 to 1 is output to ensure that the scoring range is uniform and controllable.

[0157] S440. Calculate the session consistency score for each knowledge block in the language-tagged knowledge block set based on the currently locked terms.

[0158] In this embodiment, the proportion of matching and non-matching terms in the knowledge block is statistically analyzed to quantify the coherence between the content and the contextual terminology system. A neutral score is used when there are no locked terms in the first round.

[0159] S450. Calculate the score of each knowledge block in the language tag knowledge block set based on the preset scoring function, semantic similarity, term alignment score, and conversation consistency score.

[0160] In this embodiment, the method uses fixed default weights and supports adaptive adjustment within intervals. The calculation formula is as follows: ; Where sim(query,chunk)∈[0,1]: semantic similarity from the embedding model; termAlign(chunk)∈[-1,1]: Terminology alignment score; consistency(chunk)∈[0,1]: consistency with previous conversation terms; α, β, γ are weights (default: α=0.5, β=0.3, γ=0.2); The query refers to the text currently entered by the user, that is, the question or search statement posed by the user; A chunk refers to a single bilingual knowledge block, which is an independent knowledge point content unit stored in the bilingual block index.

[0161] In this embodiment, the value range, default value, and meaning of each weight are as follows: Semantic weight α: ranges from [0.3, 0.7], with a default value of 0.5, representing the basic relevance given by the embedding model; Term alignment weight β: Value range [0.2, 0.5], default value 0.3, indicating the anchoring strength of the terminology; Consistency weight γ: Value range [0.1, 0.3], default value 0.2, indicating the importance of session coherence.

[0162] In this embodiment, the term alignment score (termAlign(chunk)) is calculated on a per-term basis: The baseline coefficient for standard terminology matching is 1.0. The coefficient for the alternative term matching ranges from 0.7 to 0.9; When a cross-language trap is detected, a penalty of 0.5–1.0 will be applied. The raw scores are calculated by summing all individual coefficients, and then normalized using score / max(term_count,1) to obtain standardized term-aligned scores.

[0163] In this embodiment, the session consistency score (consistency(chunk)) is used to check whether the terms used in the current knowledge chunk are consistent with the terms locked in the previous conversation.

[0164] If no terms have been locked before, the intermediate score of 0.5 is returned directly. If there are already locked terms, count the number of matching terms (those that used locked terms) and the number of non-matching terms (those that used other valid terms). Finally, the consistency score is calculated by dividing the number of matches by the sum of the number of matches and the number of mismatches (i.e., using the formula matches / (matches+mismatches)).

[0165] S460. Reorder the set of language-marked knowledge blocks according to the scoring scores to obtain a reordered list of knowledge blocks.

[0166] In this embodiment, during the reordering process, when the score difference between multiple knowledge blocks is within a preset score threshold, a deterministic tie-breaking rule is used to determine the order. The order priority determined by the deterministic tie-breaking rule is as follows: alignment confidence, terminology standardization score, knowledge block length, and knowledge block identifier.

[0167] To improve the adaptability and usability of the query content while ensuring the accuracy and standardization of the results, step S460 may also include: S461. Reorder the set of language-marked knowledge blocks according to the scoring scores to obtain an initial sorted list.

[0168] In this embodiment, the initial sorting is completed based on the comprehensive score from high to low, and highly relevant and compliant knowledge content is selected first.

[0169] S462. Based on the preset mastery level control retrieval strategy, filter the initial sorted list by block type to obtain a reordered knowledge block list.

[0170] In this embodiment, different types of knowledge blocks, such as definition blocks, example blocks, and prompt blocks, are selected based on the learner's level of understanding of the concepts to adapt to the comprehension needs of different learning stages.

[0171] For example, to ensure the reproducibility of the reordering results, when knowledge blocks have the same score, ties are broken in the following order of priority: Prioritize knowledge blocks with higher alignment confidence; When alignment confidence is the same, select the knowledge block with the higher term canonical score. When terminology scores are the same, the shorter knowledge block is selected. If the lengths are still the same, they are sorted according to the lexicographical order of the block identifiers (i.e., the lexicographical identifier ID takes precedence).

[0172] In the above embodiments, the method can comprehensively score semantic similarity, term alignment effect and conversation consistency in multiple dimensions, and ensure stable and reproducible ranking by breaking the rules through deterministic tie, so that the ranking of knowledge blocks is more in line with professional terminology norms and contextual coherence, and significantly improves the accuracy and orderliness of query results.

[0173] In some embodiments, step S500 may include: S510. Obtain the first knowledge block ranked first from the reordered knowledge block list, and obtain the concept terms associated with the first knowledge block.

[0174] In this embodiment, the top knowledge block with the best overall performance is selected, and the core professional concepts it is bound to are extracted as candidate objects for terminology locking.

[0175] S520. When the current session state does not include the concept term associated with the first knowledge block, the concept term associated with the first knowledge block is identified as the new locked term.

[0176] In this embodiment, terminology locking is only performed on newly added concepts that are not fixed within the session to avoid frequent changes to the standard expression.

[0177] S530. Update the current session state according to the new locking term to obtain the new session state.

[0178] In this embodiment, newly added locked terms are written into the session state to achieve session-level term persistence, while also supporting extended implementation methods for multi-session persistent locking by course or module.

[0179] For example, this embodiment employs a terminology locking rule to avoid fluctuations in terminology within a session. The specific strategy is as follows: When a conceptual term appears for the first time, the term used in the highest-ranked knowledge block is used to lock it. Once a term is locked, it remains valid throughout the session and will only be unlocked upon receiving an explicit unlock instruction. In normal mode, the locking relationship only takes effect within the current session and is not retained across sessions.

[0180] In this embodiment, in an alternative implementation for educational scenarios, if it is necessary to maintain consistency of terminology across sessions, terminology locking can be persistently managed based on courses, modules, or specified time periods.

[0181] In the above embodiments, the method can dynamically filter high-quality core concept terms based on the knowledge block sorting results, complete the iterative update of session-locked terms, realize the continuous optimization and solidification of the state of session context terms, ensure the uniformity and continuity of term usage within the same session, avoid term conflicts and expression confusion in the context, and continuously improve the standardization and consistency of cross-language query responses.

[0182] Based on the implementation of the above method, this application also provides various measurement methods to evaluate system performance. These various measurement methods include: Session log analysis is used to measure terminology consistency rate, which reflects the percentage of rounds in which locked canonical terms are retained in the session; Cross-language Recall@k is measured using a bilingual test set to evaluate the ability of Chinese queries to retrieve the correct English chunks in the top-k results; The trap avoidance rate is measured using a trap pattern test set, which is used to statistically analyze the proportion of false cognate cases that are correctly detected and corrected. Term lock stability is measured by analyzing session logs, which reflects the percentage of sessions in which no non-corrective term changes occur. The accuracy of language selection is measured by comparing expert annotations, which reflects the degree of consistency between the system's language selection and expert recommendations. Reordering latency is measured through performance benchmarks to characterize the P95 time of reordering under glossary constraints (where P95 refers to the response time of the 95th percentile of all interfaces / programs over a period of time, sorted from smallest to largest).

[0183] Based on this, a comprehensive test comparison was conducted using the aforementioned multi-dimensional evaluation indicators. Compared to the traditional baseline scheme that does not introduce glossary constraints, this method has the following technical advantages: By relying on the terminology locking mechanism, the problem of expression confusion caused by embedding vector drift is effectively improved, and the degree of terminology consistency throughout the conversation is greatly enhanced; By relying on trap detection capabilities, it can make up for the omissions in simple semantic similarity retrieval and effectively avoid cognitive errors caused by cross-linguistic pseudo-cognates; By combining concept anchoring design, the limitations of pure embedded retrieval are optimized, and the cross-language retrieval accuracy of professional and technical terms is comprehensively enhanced. By adopting an adaptive language selection strategy to replace the traditional random or fixed configuration mode, the rationality of language selection in bilingual teaching scenarios is significantly improved.

[0184] Meanwhile, the core functionality of this method is largely implemented using rule-based and feature-based approaches, thus enabling real-time inference without relying on Large Language Models (LLMs), as detailed below: Terminology matching is achieved through precise string matching and word segmentation, without relying on LLM; The terminology graph traversal is based on graph algorithms such as BFS and DFS, and does not depend on LLM; Trap pattern detection is achieved through regular expressions and dictionary lookups, without relying on an LLM; The terminology alignment score is calculated using a weighted summation formula and is independent of LLM. Session term locking is achieved through a state machine and hash mapping, without relying on an LLM; The language selection is based on the Sigmoid function and threshold judgment, and does not depend on LLM; The re-ranking score is performed using a linear combination formula and does not depend on LLM. Concept ID filtering is implemented through index lookup and filtering, and does not depend on LLM; Embedsion similarity is calculated using vector dot product, and related results are obtained through offline pre-computation.

[0185] Therefore, it can be seen that among the 10 core operations mentioned above, 9 are pure rule-based implementations, with only pre-computed vectors used for embedding similarity. This architecture enables the invention to possess stable characteristics of being interpretable, reproducible, and engineerable, rather than simply relying on AI black-box decision-making.

[0186] like Figure 2 As shown, some embodiments of this application provide a schematic diagram of the structure of a data query device based on term anchoring. It should be understood that this device is related to... Figure 1The method executed in the middle corresponds to the steps involved in the aforementioned method. The specific functions and effects of the device can be found in the description above. To avoid repetition, detailed descriptions are omitted here.

[0187] The term-anchored data query device includes: The receiving unit 910 is used to receive query data input by the user and the current session status; The identification unit 920 is used to identify the query data and obtain a final list of concept terms and a list of language decision tags corresponding to the final list of concept terms; Unit 930 is used to obtain a set of language markup knowledge blocks based on the final list of conceptual terms and query data. The reordering unit 940 is used to reorder the language mark knowledge block set based on a scoring function according to the current session state, a preset bilingual terminology list, and query data to obtain a reordered knowledge block list; wherein, the standard term matching result and the alternative term matching result used in the scoring function are score-boosting items, and the cross-language trap detection result is a score-penalizing item. Update unit 950 is used to update the locked terms of the current session state based on the reordered knowledge block list to obtain the new session state; Output unit 960 is used to output query answers based on the final list of conceptual terms, the list of language decision tags, the list of reordered knowledge blocks, and the locked terms in the new session state.

[0188] In some embodiments, the identification unit 920 includes: The identification subunit 921 is used to identify query concepts and terms in the query data to obtain a final list of concepts and terms. The tagging subunit 922 is used to perform language version tagging processing on the final concept term list according to the preset bilingual terminology list to obtain the language decision tagging list; The language version can be a first language version, a second language version, or a bilingual version.

[0189] In some embodiments, the tagging subunit 922 is specifically used to obtain the user's user profile and conceptual terminology mastery level; obtain multiple difficulty ratings corresponding to multiple conceptual terms in the final conceptual terminology list according to a preset bilingual terminology list; obtain the user's second language proficiency level according to the user profile, and obtain the second language immersion coefficient according to the user profile; obtain multiple current mastery levels corresponding to multiple conceptual terms in the final conceptual terminology list according to the conceptual terminology mastery level; calculate multiple confidence levels corresponding to multiple conceptual terms in the final conceptual terminology list according to a preset first weight parameter, multiple difficulty ratings, multiple current mastery levels, second language proficiency level, and second language immersion coefficient; and perform language version tagging processing on multiple conceptual terms in the final conceptual terminology list according to multiple confidence levels and preset judgment rules to obtain a language decision tagging list. The determination rules include: Conceptual terms with a confidence level not less than a preset first confidence threshold are labeled as second languages; Conceptual terms with a confidence level not greater than a preset second confidence threshold are marked as first language conceptual terms; Conceptual terms with a confidence level less than a preset first confidence level threshold and greater than a preset second confidence level threshold are marked as bilingual conceptual terms; The first confidence threshold is greater than the second confidence threshold; The bilingual glossary maintains a terminology graph for each concept term, including nodes and edges; Nodes represent canonical terminology, alternative terminology, and trap patterns; The relationships represented by edges include translation relations, priority relations, avoidance relations, and context dependency relations; The edge weights are used in the scoring function.

[0190] In some embodiments, the acquisition unit 930 includes: The first acquisition subunit 931 is used to acquire a set of concept term identifiers based on the final concept term list; The first acquisition subunit 931 is further configured to acquire a first knowledge block set based on a preset bilingual block index and a set of concept term identifiers; wherein, the list of concept terms corresponding to each knowledge block in the first knowledge block set includes at least one concept term identifier from the set of concept term identifiers. Search subunit 932 is used to perform a similarity search in a preset bilingual block index based on the query data to obtain a second set of knowledge blocks; Merge subunit 933 is used to merge the first knowledge block set and the second knowledge block set to obtain the initial knowledge block set; Deduplication subunit 934 is used to deduplicatize the initial knowledge block set to obtain the associated knowledge block set; The tagging subunit 935 is used to tag the language version type of each knowledge block in the associated knowledge block set according to the language decision type, so as to obtain the language tagging knowledge block set.

[0191] In some embodiments, the reordering unit 940 includes: The second acquisition subunit 941 is used to acquire the current lock term based on the session state; Computational subunit 942 is used to calculate the semantic similarity between each knowledge block in the set of language-tagged knowledge blocks and the query data; The calculation subunit 942 is also used to calculate the term alignment score of each knowledge block in the language-tagged knowledge block set according to the preset bilingual terminology list; The computational subunit 942 is also used to calculate the session consistency score of each knowledge block in the language-tagged knowledge block set based on the currently locked terminology; The computational subunit 942 is also used to calculate the score of each knowledge block in the language-marked knowledge block set based on a preset scoring function, semantic similarity, term alignment score and conversation consistency score; The sorting subunit 943 is used to reorder the set of language-marked knowledge blocks according to the score, and obtain a reordered list of knowledge blocks. In the re-ranking process, when the score difference between multiple knowledge blocks is within a preset score threshold, a deterministic tie-breaking rule is used to determine the ranking. The ranking priority determined by the deterministic tie-breaking rule is, in order, alignment confidence, terminology standardization score, knowledge block length, and knowledge block identifier.

[0192] In some embodiments, the sorting subunit 943 is specifically used to reorder the set of language-marked knowledge blocks according to the rating score to obtain an initial sorted list; and to filter the initial sorted list by block type according to a preset mastery control retrieval strategy to obtain a reordered knowledge block list.

[0193] In some embodiments, calculating the term alignment score for each knowledge block in the language-tagged knowledge block set based on a preset bilingual terminology list includes: Retrieve a list of multiple concept terms that correspond one-to-one with multiple knowledge blocks from the language-marked knowledge block set; The pre-defined bilingual terminology list is matched one by one with multiple lists of conceptual terms to obtain matching results; among them, the matching results include standard term matching results, alternative term matching results, cross-language trap detection results, and no matching results; Based on the matching results, determine multiple lists of matching results that correspond one-to-one with multiple lists of concept terms; Calculate the term alignment score in the language tag knowledge block set that corresponds one-to-one with multiple knowledge blocks based on the multiple matching result assignment lists; Among them, the assigned values ​​corresponding to the standard term matching results and the alternative term matching results are score boost values, and the magnitude result corresponding to the cross-language trap detection results is the score penalty value.

[0194] In some embodiments, the term-anchored data query apparatus further includes: The acquisition unit 930 is also used to acquire cross-language trap patterns based on the bilingual terminology list; wherein, the cross-language trap patterns are direction-sensitive, that is, the trap patterns from the first language to the second language and the trap patterns from the second language to the first language are directionally related; Misunderstanding detection unit 970 is used to perform misunderstanding detection on query data based on cross-language trap patterns, obtain a list of misunderstandings and output it. The list of misunderstandings includes results that are not misunderstandings or results that are misunderstandings. The existence of a misunderstanding includes at least one type of misunderstanding and the severity level corresponding to each type of misunderstanding; Misunderstanding types include one or more of the following: pseudo-cognates, literal translation errors, terminology confusion, conceptual term transfer errors, and cross-linguistic symbolic convention differences.

[0195] In some embodiments, the update unit 950 includes: The third acquisition subunit 951 is used to acquire the first knowledge block ranked first from the reordered knowledge block list, and to acquire the concept terms associated with the first knowledge block. Determine subunit 952, used to determine the concept term associated with the first knowledge block as a new locked term when the current session state does not include the concept term associated with the first knowledge block; Update subunit 953 is used to update the current session state according to the new locking term to obtain the new session state.

[0196] like Figure 3 As shown, this application provides an electronic device 1000, which includes a processor 1001 and a memory 1002. The processor 1001 and the memory 1002 are interconnected and communicate with each other through a communication bus 1003 and / or other forms of connection mechanism (not shown). The memory 1002 stores a computer program that can be executed by the processor 1001. When the computing device is running, the processor 1001 executes the computer program to perform the method in any of the aforementioned optional implementations.

[0197] This application provides a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the method in any of the aforementioned optional implementations.

[0198] The computer-readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0199] This application provides a computer program product, which includes a computer program that, when run by a processor, executes the method in any of the aforementioned optional implementations.

[0200] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and not to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. These modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application, and they should all be covered within the scope of the claims and specification of this application. In particular, as long as there is no conflict, the various technical features mentioned in the embodiments can be combined in any way. This application is not limited to the specific embodiments disclosed herein, but includes all technical solutions falling within the scope of the claims.

Claims

1. A data query method based on term anchoring, characterized in that, include: Receive query data and current session status input by the user; The query data is identified to obtain a final list of conceptual terms and a list of language decision tags corresponding to the final list of conceptual terms. Based on the final list of conceptual terms and the query data, obtain a set of language-marked knowledge blocks; Based on the current session state, the preset bilingual terminology list, and the query data, the language-marked knowledge block set is reordered using a scoring function to obtain a reordered knowledge block list; wherein, the standard term matching result and the alternative term matching result used in the scoring function are score-boosting items, and the cross-language trap detection result is a score-penalizing item; The current session state is updated with locked terms based on the reordered knowledge block list to obtain a new session state. The query answer is output based on the final list of conceptual terms, the list of language decision tags, the list of reordered knowledge blocks, and the locked terms in the new session state.

2. The data query method based on term anchoring according to claim 1, characterized in that, The process of identifying the query data to obtain a final list of concept terms and a list of language decision tags corresponding to the final list of concept terms includes: The query data is subjected to query concept term identification to obtain a final list of concept terms; The final conceptual terminology list is processed by language version tagging based on a preset bilingual terminology list to obtain a language decision tagging list; The language version is one of a first language version, a second language version, or a bilingual version.

3. The data query method based on term anchoring according to claim 2, characterized in that, The step of performing language version tagging processing on the final concept term list according to the preset bilingual terminology list to obtain a language decision tag list includes: Obtain the user profile and conceptual terminology mastery level of the user; Based on the preset bilingual terminology list, obtain multiple difficulty ratings that correspond one-to-one with multiple concept terms in the final concept terminology list; The user's second language proficiency level is obtained based on the user profile, and the second language immersion coefficient is obtained based on the user profile. Based on the mastery level of the concept terms, obtain multiple current mastery levels corresponding one-to-one with multiple concept terms in the final list of concept terms; Based on the preset first weight parameter, the multiple difficulty ratings, the multiple current mastery levels, the second language proficiency level, and the second language immersion coefficient, calculate multiple confidence levels that correspond one-to-one with the multiple concept terms in the final list of concept terms; Based on the multiple confidence levels and preset judgment rules, language version tagging is performed on multiple concept terms in the final concept term list to obtain a language decision tag list; The determination rules include: Conceptual terms with a confidence level not less than a preset first confidence threshold are labeled as second languages; Conceptual terms with a confidence level not greater than a preset second confidence threshold are marked as first language conceptual terms; Conceptual terms with a confidence level less than the preset first confidence level threshold and greater than the preset second confidence level threshold are marked as bilingual conceptual terms; The first confidence threshold is greater than the second confidence threshold; The bilingual terminology maintains a terminology graph for each conceptual term, including nodes and edges; The nodes represent canonical terms, alternative terms, and trap patterns; The relationships represented by the edges include translation relationships, priority relationships, avoidance relationships, and context dependency relationships; The weights of the edges are used in the scoring function.

4. The data query method based on term anchoring according to claim 1, characterized in that, The step of obtaining a set of language-marked knowledge blocks based on the final list of conceptual terms and the query data includes: Obtain the set of concept term identifiers based on the final list of concept terms; A first knowledge block set is obtained based on a preset bilingual block index and the set of concept term identifiers; wherein, the list of concept terms corresponding to each knowledge block in the first knowledge block set includes at least one concept term identifier from the set of concept term identifiers; Based on the query data, a similarity search is performed in the preset bilingual block index to obtain the second knowledge block set; Merge the first knowledge block set and the second knowledge block set to obtain the initial knowledge block set; The initial set of knowledge blocks is deduplicated to obtain a set of associated knowledge blocks; The language version type of each knowledge block in the associated knowledge block set is marked according to the language decision type to obtain the language-marked knowledge block set.

5. The data query method based on term anchoring according to claim 1, characterized in that, The step involves reordering the language-tagged knowledge block set based on a scoring function according to the current session state, a preset bilingual terminology list, and the query data, resulting in a reordered knowledge block list, including: Obtain the current lock term based on the session state; Calculate the semantic similarity between each knowledge block in the language-tagged knowledge block set and the query data; The terminology alignment score of each knowledge block in the language-marked knowledge block set is calculated based on a preset bilingual terminology list. Calculate the session consistency score for each knowledge block in the language-tagged knowledge block set based on the currently locked terminology; The score of each knowledge block in the language-marked knowledge block set is calculated based on the preset scoring function, the semantic similarity, the term alignment score, and the conversation consistency score. The language-marked knowledge block set is reordered based on the score to obtain a reordered knowledge block list. In the reordering process, when the score difference between multiple knowledge blocks is within a preset score threshold, a deterministic tie-breaking rule is used to determine the order. The order priority determined by the deterministic tie-breaking rule is as follows: alignment confidence, terminology standardization score, knowledge block length, and knowledge block identifier.

6. The data query method based on term anchoring according to claim 5, characterized in that, The language-marked knowledge block set is reordered based on the scoring scores to obtain a reordered knowledge block list, including: The language-marked knowledge block set is reordered based on the scoring scores to obtain an initial sorted list; Based on a preset mastery control retrieval strategy, the initial sorted list is filtered by block type to obtain a reordered knowledge block list.

7. The data query method based on term anchoring according to claim 5, characterized in that, The step of calculating the term alignment score for each knowledge block in the language-tagged knowledge block set based on a preset bilingual terminology list includes: Obtain a list of multiple concept terms that correspond one-to-one with multiple knowledge blocks in the language-marked knowledge block set; The preset bilingual terminology list is matched one by one with the multiple lists of conceptual terms to obtain matching results; wherein, the matching results include standard term matching results, alternative term matching results, cross-language trap detection results, and no matching results; Based on the matching results, determine multiple matching result assignment lists that correspond one-to-one with the multiple lists of concept terms; Calculate the term alignment score in the language tag knowledge block set that corresponds one-to-one with the multiple knowledge blocks according to the multiple matching result assignment lists; Among them, the assignment results corresponding to the standard term matching results and the alternative term matching results are score boosting values, and the amplitude result corresponding to the cross-language trap detection results is a score penalty value.

8. The data query method based on term anchoring according to claim 1, characterized in that, The method further includes: Cross-language trap patterns are obtained based on the bilingual terminology list; wherein, the cross-language trap patterns are direction-sensitive, and the trap patterns from the first language to the second language and the trap patterns from the second language to the first language are directionally related; The query data is misunderstanding detected based on the cross-language trap pattern, and a list of misunderstandings is obtained and output. The misunderstanding list includes results where there is no misunderstanding or results where there is a misunderstanding; The existence of a misunderstanding includes at least one type of misunderstanding and a severity level corresponding to each type of misunderstanding; The types of misunderstandings include one or more of the following: pseudo-cognates, literal translation errors, terminology confusion, conceptual terminology transfer errors, and cross-linguistic symbolic convention differences.

9. The data query method based on term anchoring according to claim 1, characterized in that, The step of updating the locked terms in the current session state according to the reordered knowledge block list to obtain the new session state includes: Obtain the first knowledge block that is ranked first from the reordered knowledge block list, and obtain the concept term associated with the first knowledge block; When the current session state does not include the concept term associated with the first knowledge block, the concept term associated with the first knowledge block is identified as a new locked term; The current session state is updated according to the new locking term to obtain the new session state.

10. A data query device based on term anchoring, characterized in that, The term-anchored data query device includes: The receiving unit is used to receive query data input by the user and the current session status; The identification unit is used to identify the query data to obtain a final list of concept terms and a list of language decision tags corresponding to the final list of concept terms; The acquisition unit is used to acquire a set of language mark knowledge blocks based on the final list of conceptual terms and the query data; The reordering unit is used to reorder the language-marked knowledge block set based on a scoring function according to the current session state, a preset bilingual terminology list, and the query data to obtain a reordered knowledge block list; wherein, the standard term matching result and the alternative term matching result used in the scoring function are score-boosting items, and the cross-language trap detection result is a score-penalizing item. The update unit is used to update the current session state by locking terms according to the reordered knowledge block list to obtain a new session state; The output unit is used to output the query answer based on the final concept term list, the language decision tag list, the reordered knowledge block list, and the locked terms in the new session state.

Images