A data automatic checking method and system based on difference detection and intelligent matching

Abstract

The application relates to the technical field of language learning evaluation, and discloses a data automatic checking method and system based on difference detection and intelligent matching, which comprises the following steps: obtaining user grades, region information and to-be-detected texts, utilizing a difference detection operator to compare a preset standard database to identify a set of words beyond the curriculum; subsequently, context features of the words beyond the curriculum are extracted, and a space mapping algorithm is utilized to generate context semantic anchor point vectors; meanwhile, historical sample individual features and benchmark features are fused to generate a current effective fingerprint; the system further extracts logical connection features of the to-be-detected texts, calculates manifold distances of the to-be-detected texts and the current effective fingerprint to determine non-original deviation risks, calculates a cognitive confirmation operator according to a correct rate and a time length deviation of user answers; and finally, target texts consistent with the cognitive level of the user are output. The application constructs a closed-loop adjustment mechanism through behavior feedback and style deviation degree, realizes accurate detection of individual cognitive boundaries and self-adaptive rollback adjustment of text styles.

Description

Technical Field

[0001] This invention relates to the field of language learning assessment technology, specifically to an automatic data verification method and system based on difference detection and intelligent matching. Background Technology

[0002] In academic writing and language learning, automated editing tools and text quality detection systems are increasingly widely used. Current text processing technologies mainly rely on large-scale language models or pre-set grammar rule bases. By identifying spelling errors, grammatical defects, and inappropriate word collocations in the text, they provide targeted optimization suggestions. In order to achieve higher quality language expression, they usually tend to introduce advanced vocabulary and complex syntactic structures to improve the smoothness of the text's logical connection and literary expression.

[0003] However, when using existing technologies, automated editing algorithms often lack effective modeling of users' true cognitive abilities. Although the generated text may be more aesthetically pleasing, it can easily lead to a significant deviation of the text features from the user's actual writing level in terms of statistical distribution, resulting in a non-original bias in the output. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an automatic data verification method and system based on difference detection and intelligent matching, which solves the problem of non-original offset characteristics in the output results.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an automatic data verification method based on difference detection and intelligent matching, comprising the following steps:

[0006] Obtain the grade and region information associated with the user, and obtain the text to be detected uploaded by the user. By comparing it with the preset standard database, identify and output a set of words that are beyond the scope of the curriculum.

[0007] Extract the contextual features of the over-the-top word set in the text to be detected, and generate the corresponding contextual semantic anchor vector;

[0008] The individual feature vectors extracted from historical samples are combined with the baseline feature vectors extracted from the standard database to generate the current valid fingerprint;

[0009] Calculate the distance between the logical connection features of the text to be detected and the current valid fingerprint to determine the risk of non-original deviation;

[0010] The question is verified based on the contextual semantic anchor vector matching, and the cognitive confirmation operator is calculated based on the user's answer response data;

[0011] The polished text is generated within a probability field defined by the current valid fingerprint and the set of out-of-scope words. The style regression adjustment is performed using the cognitive confirmation operator and the non-original deviation risk to produce the target text.

[0012] Preferably, obtaining the user-uploaded text to be detected includes the following steps:

[0013] The system receives raw text uploaded by users through an interactive interface and uses character cleaning operators to remove non-target language characters from the raw text to obtain standardized text.

[0014] The standardized text is subjected to language detection and legality verification. If the verification passes, the standardized text is input as the text to be detected into subsequent steps.

[0015] The preset standard database is a syllabus vocabulary database set up based on the grade and region information.

[0016] Preferably, the identification and output of the over-the-top word set includes the following steps:

[0017] Lexical features are extracted from the text to be detected by a difference detection operator and compared with the vector cosine similarity of the compliant vocabulary in the preset standard database.

[0018] Words not covered by the compliant thesaurus are marked as out-of-scope candidate words, and the out-of-scope candidate words are filtered and deduplicated according to the syntactic structure of the text to be detected to generate the out-of-scope word set.

[0019] Preferably, the extraction of contextual features of the over-the-top word set in the text to be detected includes the following steps:

[0020] Using a sliding window operator, context-related terms within a preset span are obtained, centered on each super-category word in the super-category word set.

[0021] The context-related terms are vector-encoded using a pre-trained language model, and combined with the position index of the super-class words in the text to be detected, contextual features representing the semantic environment of the super-class words are calculated.

[0022] Preferably, the step of generating the corresponding context semantic anchor vector is as follows:

[0023] By assigning weights to the contextual features using contextual semantic operators, the key nodes that contribute the most to the semantic orientation of the super-category word set are identified.

[0024] The key nodes are converted into contextual semantic anchor vectors with specific coordinate dimensions in the feature space using a spatial mapping algorithm.

[0025] Preferably, generating a currently valid fingerprint includes the following steps:

[0026] The weighting coefficients corresponding to the sample throughput of the historical samples are calculated using the confidence smoothing function;

[0027] Based on the weighting coefficients, the individual feature vector and the baseline feature vector are linearly or non-linearly weighted and fused to generate a current valid fingerprint representing the corresponding user's capability characteristics.

[0028] Preferably, calculating the distance between the logical connection features of the text to be detected and the current valid fingerprint includes the following steps:

[0029] The conjunction transition matrix and semantic transition probability in the text to be detected are extracted by the syntactic dependency parsing operator.

[0030] Construct a logical connection vector and use a manifold learning algorithm to calculate the manifold distance between the logical connection vector and the current valid fingerprint in the feature space.

[0031] Preferably, the step of calculating the cognitive confirmation operator based on the user's answer response data includes the following steps:

[0032] The cognitive load analysis operator is used to extract the user's correct answer rate and response time for the verification questions, and the deviation between the response time and the preset standard time is calculated.

[0033] The accuracy rate of answering questions and the deviation rate are coupled and calculated to generate a cognitive confirmation operator that represents the user's certainty in mastering advanced vocabulary.

[0034] Preferably, the style regression adjustment using the cognitive confirmation operator includes the following steps:

[0035] The quantized value of the cognitive confirmation operator is obtained through an adaptive backoff regulator;

[0036] If the value is lower than a preset threshold or the risk of non-original deviation exceeds a preset threshold, the smoothness of logical connection of the polished text is reduced according to the inverse weight of the value or the risk.

[0037] By using syntactic weakening operators to reduce the structural dimensionality of sentence structures with a preset complexity or higher in the polished text, the final target text is reduced in terms of logical complexity and pragmatic style to the cognitive range represented by the current effective fingerprint.

[0038] Preferably, an automatic data verification system based on difference detection and intelligent matching includes the following modules:

[0039] The difference detection and out-of-syllabus recognition module is used to acquire grade and region information and the text to be detected. It uses the difference detection operator to compare the lexical features of the text with the vector cosine similarity of the preset standard database, and identifies and outputs a set of out-of-syllabus words.

[0040] Contextual semantic anchor generation module: used to extract contextual features of the over-the-top word set, identify key nodes through contextual semantic operators, and generate corresponding contextual semantic anchor vectors using spatial mapping algorithms;

[0041] The current valid fingerprint modeling module is used to fuse individual feature vectors extracted from historical samples with baseline feature vectors extracted from a standard database, calculate weight coefficients through a confidence smoothing function, and generate the current valid fingerprint.

[0042] Non-original deviation risk determination module: It is used to extract the conjunction transformation matrix and semantic transition probability through syntactic dependency parsing operator and construct logical connection vector, and use manifold learning algorithm to calculate the distance between it and the current valid fingerprint to determine the non-original deviation risk;

[0043] Cognitive confirmation operator solution module: used to verify questions based on context semantic anchor vector matching. It extracts the answer accuracy and response time through the cognitive load analysis operator and calculates the deviation, and then couples the solution to obtain the cognitive confirmation operator.

[0044] Style regression adjustment and output module: It is used to generate polished text in a probability field. It uses an adaptive regression adjuster to adjust style regression based on cognitive confirmation operators and non-original deviation risk, and performs structural dimensionality reduction through syntactic weakening operators to output target text.

[0045] This invention provides a method and system for automatic data verification based on difference detection and intelligent matching. It has the following beneficial effects:

[0046] 1. This invention uses the cognitive confirmation operator and the risk of non-original deviation as negative feedback signals, and performs structural dimensionality reduction processing on the polished text by the syntactic weakening operator. It can dynamically prune the depth of the syntactic tree according to the feedback intensity coefficient, so that the final target text is back to the cognitive range represented by the current effective fingerprint in terms of vocabulary selection and logical complexity. This ensures that the optimized text maintains consistency with the user's actual level while improving quality.

[0047] 2. This invention utilizes contextual semantic anchor vectors to retrieve context-aligned verification questions from a pre-set question bank, and combines the user's answer accuracy and response time deviation to calculate a cognitive confirmation operator. By quantifying the individual's cognitive load through behavioral temporal data, it effectively eliminates the interference of accidental correct answers and can detect the user's true mastery of advanced vocabulary and complex grammar in real time. Thus, it automatically and objectively defines the boundaries of an individual's language ability without relying on human intervention.

[0048] 3. This invention extracts the conjunction transition matrix and semantic transition probability through syntactic dependency analysis operators, and uses manifold learning algorithms to calculate the manifold distance between the text to be detected and the current valid fingerprint. It identifies the risk of non-original deviation from the dimension of deep logical organization pattern. Compared with traditional keyword overlap retrieval, it can more accurately capture the logical connection anomalies caused by generative tools or others' polishing, and provides a scientific quantitative basis for determining the authenticity of text. Attached Figure Description

[0049] Figure 1 This is a flowchart of an automatic data verification method based on difference detection and intelligent matching according to the present invention.

[0050] Figure 2 This is a structural diagram of an automatic data verification system based on difference detection and intelligent matching according to the present invention. Detailed Implementation

[0051] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0052] Example 1

[0053] Please see the appendix Figure 1 This invention provides an automatic data verification method based on difference detection and intelligent matching, comprising the following steps:

[0054] Obtain the grade and region information associated with the user, and obtain the text to be detected uploaded by the user. By comparing it with the preset standard database, identify and output a set of words that are beyond the scope of the curriculum.

[0055] Furthermore, obtaining the user-uploaded text to be detected includes the following steps:

[0056] The system receives raw text uploaded by users through an interactive interface and uses character cleaning operators to remove non-target language characters from the raw text to obtain standardized text.

[0057] The standardized text is checked for language and legality. If the check passes, the standardized text is used as the text to be checked and then input into the subsequent steps.

[0058] The default standard database is a syllabus vocabulary set up based on grade and region information;

[0059] Identifying and outputting the set of words beyond the scope of the syllabus includes the following steps:

[0060] Lexical features are extracted from the text to be detected by the difference detection operator and compared with the vector cosine similarity of the lexicon in the preset standard database.

[0061] Words not covered by the compliant thesaurus are marked as out-of-scope candidate words, and the out-of-scope candidate words are filtered and deduplicated according to the syntactic structure of the text to be detected to generate an out-of-scope word set.

[0062] Specifically, the system first obtains the user's associated grade and region information, and then receives the raw text uploaded by the user through the system interface. A character cleaning operator is used to remove non-target language characters, thus transforming the raw data into standardized text. After passing language detection and legality verification, this standardized text is used as input for detection. During difference detection, distance distribution in a high-dimensional feature space is used to identify outliers. Each word in the text to be detected is mapped to a word feature vector with fixed dimensions through a pre-trained model. Simultaneously, the standard vocabulary in the corresponding teaching syllabus vocabulary database is clustered and mapped to compliant vocabulary center vectors. By calculating the cosine similarity between the two:

[0063] ;

[0064] This measure assesses the degree of deviation of a word from the outline requirements. The similarity value ranges from 0 to 1, with values ​​closer to 1 indicating a closer resemblance in pragmatic features to the outline content. A similarity threshold, dynamically generated based on the outline's difficulty level, is also used. This threshold is typically between 0.7 and 0.9. If the calculated similarity... Less than this threshold If the word is not found to be a candidate word, it is identified as an out-of-syllabus word. To eliminate interference caused by part-of-speech transformation or proper nouns, the syntactic analysis operator is further invoked to analyze the components of the candidate word in the sentence, retaining only the content words that serve as core semantic support. After filtering and deduplication, the final set of out-of-syllabus words is produced. By combining the abstract vector comparison algorithm with the feature analysis of the lexicon in the field of education, the system is able to automatically anchor the user's ability boundary across different regions, providing accurate data parameter support for subsequent steps.

[0065] Extract the contextual features of the over-the-top word set in the text to be detected, and generate the corresponding contextual semantic anchor vector;

[0066] Furthermore, extracting contextual features of the over-the-top word set in the text to be detected includes the following steps:

[0067] Using the sliding window operator, context-related terms within a preset span are obtained, centered on each super-category word in the super-category word set.

[0068] By using a pre-trained language model to encode context-related terms and combining the position index of the over-the-top words in the text to be detected, context features representing the semantic environment of the over-the-top words are calculated.

[0069] The steps to generate the corresponding context semantic anchor vector are as follows:

[0070] By assigning weights to contextual features using contextual semantic operators, the key nodes that contribute the most to the semantic orientation of the over-the-top word set are identified.

[0071] The key nodes are transformed into contextual semantic anchor vectors with specific coordinate dimensions in the feature space using a spatial mapping algorithm;

[0072] Specifically, to capture the semantic coupling strength between out-of-scope words and their neighboring words in order to eliminate the interference of word polysemy, a sliding window is first used to obtain contextually related terms within a span centered on each out-of-scope word in the out-of-scope word set, extending to both sides. This preset span is usually set to three to five words. Then, the semantic model of the pre-trained model is used to encode these related terms into vectors, and combined with the position index of the out-of-scope word in the text to be detected, a contextual feature vector representing the context of the out-of-scope word is calculated. Based on this, attention weights are assigned to contextual features using contextual semantic operators. The calculation logic is as follows:

[0073] ;

[0074] in, The query vector represents the core semantics of words beyond the syllabus. As a key vector representing context-related terms, The weight matrix is ​​calculated using a preset scaling factor, typically the square root of the feature dimension. The statistical contribution of quantified context words to the semantic meaning of words beyond the syllabus is used to identify the key nodes with the highest contribution. Finally, a spatial mapping algorithm is used to transform these key nodes through a linear transformation matrix. Transformed into context semantic anchor vectors with specific coordinate positions in the feature space. The calculation formula is:

[0075] ;

[0076] in, The feature vectors of key nodes are used to ensure the accuracy of cognitive verification results by fusing isolated out-of-syllabus words with their specific linguistic context.

[0077] The individual feature vectors extracted from historical samples are combined with the baseline feature vectors extracted from the standard database to generate the current valid fingerprint;

[0078] Furthermore, generating a currently valid fingerprint includes the following steps:

[0079] The weighting coefficients corresponding to the sample throughput of historical samples are calculated using the confidence smoothing function;

[0080] Based on the weighting coefficients, the individual feature vector and the baseline feature vector are linearly or non-linearly weighted and fused to generate the current valid fingerprint representing the corresponding user's ability characteristics;

[0081] Specifically, the system first retrieves the user's past writing texts from the historical sample database, and then uses feature extraction operators to transform them into individual feature vectors that include lexical richness, syntactic depth, and high-frequency word distribution dimensions. Simultaneously, it retrieves benchmark feature vectors from the standard database that meet the teaching syllabus requirements of the user's grade and region. By using dynamic weights to balance the information bias between individual features and the group baseline, feature center of gravity shifts under different data throughputs, and using a confidence smoothing function:

[0082] ;

[0083] Calculate the weighting coefficients, where, The number of historical samples. A preset convergence constant, typically ranging from 0.1 to 0.3 based on the desired system learning rate, is used to linearly weight and fuse the individual feature vector with the baseline feature vector, generating a currently valid fingerprint representing the corresponding user's capability characteristics. The calculation formula is as follows:

[0084] ;

[0085] Through this calculation process, when the number of samples... When the weighting coefficient is small, Approaching 0, the system fingerprint represents the outline benchmark. The main focus is on the number of sample articles. The increase in weighting coefficient Gradually approaching 1, the system fingerprint automatically smooths to match the user's personalized features. This implementation method effectively solves the problem of feature modeling failure caused by the sparsity of initial user data. Through an adaptive weight evolution mechanism, it provides a statistically stable benchmark for subsequent steps to determine whether the detected text deviates from the user's true writing level.

[0086] Calculate the distance between the logical connection features of the text to be detected and the current valid fingerprint to determine the risk of non-original deviation;

[0087] Furthermore, calculating the distance between the logical connection features of the text to be detected and the current valid fingerprint includes the following steps:

[0088] The conjunction transition matrix and semantic transition probability in the text to be detected are extracted by the syntactic dependency parsing operator.

[0089] Construct a logical connection vector and use a manifold learning algorithm to calculate the manifold distance between the logical connection vector and the current valid fingerprint in the feature space;

[0090] Specifically, the syntactic dependency parsing operator is used to perform deep analysis of the text to be tested, identify the inter-sentence connection paths, and extract the conjunction transformation matrix. This matrix records the switching frequency of words with different logical relationships in the text, and extracts the semantic transition probability, which represents the semantic jump magnitude between sentences. This leads to the construction of a logical connection vector. The principle behind determining the risk of non-originality lies in measuring the topological similarity between the text to be detected and the user's existing writing feature manifold in terms of logical evolution features. This is achieved by using a manifold learning algorithm to connect the logical connection vectors... The currently valid fingerprint generated above Map to a low-dimensional manifold space and compute the manifold distance between them:

[0091] ;

[0092] in, The feature contribution weights are automatically assigned based on the importance of the feature dimensions, and In this calculation process, manifold distance It reflects the degree of deviation between the text's logical features and the user's actual cognitive level. If the distance exceeds a dynamic threshold based on the statistical distribution of the user's historical performance, the text is judged to have a risk of non-original deviation, and the corresponding risk level is output. This implementation method uses the manifold distance metric algorithm to deeply mine the syntactic organization pattern inside the text, effectively identifying the logical connection oversmoothness or semantic leap anomaly that may be caused by generative artificial intelligence or ghostwriting.

[0093] The questions are verified based on contextual semantic anchor vector matching, and the cognitive confirmation operator is calculated based on the user's answer response data.

[0094] Furthermore, the cognitive confirmation operator is calculated based on user response data, including the following steps:

[0095] The cognitive load analysis operator extracts the user's correct answer rate and response time for the verification questions, and calculates the deviation between the response time and the preset standard time.

[0096] The correct answer rate and the deviation rate are coupled and calculated to generate a cognitive confirmation operator that represents the user’s certainty in mastering advanced vocabulary.

[0097] Specifically, firstly, based on the aforementioned generated contextual semantic anchor vector... The cosine similarity is calculated between the contextual representation vectors of each question in the preset question bank and the verification question with the highest similarity and logical alignment with the context is retrieved and pushed to the user. Subsequently, the user's response data for the verification question is received. In this embodiment, the principle of solving the cognitive confirmation operator lies in combining behavioral temporal analysis and accuracy to quantify an individual's cognitive load, thereby eliminating the interference of random factors and truly verifying the user's ability to handle features beyond the scope of the subject. The user's accuracy is verified by utilizing cognitive load analysis. Among them, accuracy The value can be 0 or 1, and the user's response time is also extracted. The standard answering time for this verification question is pre-set. Through the formula:

[0098] ;

[0099] Calculate the response deviation Based on this, the solution formula is coupled:

[0100] ;

[0101] Generate cognitive confirmation operator ,in, The preset cognitive sensitivity adjustment coefficient is used to adjust the system's sensitivity to time fluctuations. Its value is usually set between 0.5 and 1.5. Based on this mapping logic, the final cognitive confirmation operator is generated. The value range is limited to 0 to 1. If the user answers correctly and the response time is close to the standard value, the response deviation is... Approaching 0, cognitive confirmation operator A value close to 1 indicates that the user has a definite grasp of the advanced vocabulary. If the user answers incorrectly, or answers correctly but the response time deviates significantly from the standard value, the cognitive confirmation operator will be flagged. The value will decrease significantly.

[0102] The text is polished within a probability field defined by the current valid fingerprints and the set of out-of-scope words. The text is then adjusted for style regression using cognitive confirmation operators and the risk of non-original deviations, resulting in the target text.

[0103] Furthermore, style regression adjustment using cognitive confirmation operators includes the following steps:

[0104] The quantized value of the cognitive confirmation operator is obtained through an adaptive backoff regulator;

[0105] If the value is lower than the preset threshold or the risk of non-original deviation exceeds the preset threshold, the smoothness of the logical connection of the polished text will be reduced according to the inverse weight of the value or risk.

[0106] By using syntactic weakening operators to reduce the structural dimensionality of sentence structures with a preset complexity in the polished text, the final target text is reduced to the cognitive range represented by the current effective fingerprint in terms of logical complexity and pragmatic style.

[0107] Specifically, firstly, utilize currently valid fingerprints. A dynamic cognitive probability field is constructed using a super-frequency set. The principle of this probability field is to limit the sampling probability of highly difficult words and complex sentence structures during generation, thereby confining the solution space of text generation within the user's true cognitive boundary. In this embodiment, under the constraints of the above probability field, a language model is invoked to generate polished text, and then an adaptive backoff regulator obtains the cognitive confirmation operator. In this embodiment, the principle of style regression adjustment lies in constructing a feedback adjustment loop based on the user's external behavior and internal style deviation. When the polished text features are detected to significantly exceed the user's actual level, style restoration is achieved by reducing syntactic complexity. Specifically, the cognitive confirmation operator is used... Risk of deviation from original content Perform coupling and calculate the backoff strength coefficient:

[0108] ;

[0109] in, To control the steepness factor of the adjustment sensitivity, its value is usually between 2 and 5. and These are preset risk thresholds and cognitive thresholds, respectively. In this embodiment, when the backoff intensity coefficient... When the preset adjustment boundary is exceeded, the syntactic weakening operator is invoked to reduce the dimensionality of sentence structures with a complexity exceeding the preset limit in the polished text. Specifically, this involves calculating the maximum depth of the syntactic tree in the polished text. Baseline depth corresponding to the current valid fingerprint The difference And based on the retreat strength coefficient By pruning or splitting nested clauses or modifiers in the syntax tree according to a certain proportion, the final target text is constrained in terms of syntactic complexity and applicability within the cognitive range represented by the current valid fingerprint. This implementation method solves the technical defect that automated polishing algorithms are prone to exceeding the statistical limits of text features by establishing a closed-loop feature feedback mechanism, and ensures the logical consistency between the target text and the user's real ability level.

[0110] Example 2

[0111] Please see the appendix Figure 2 This invention provides an automatic data verification system based on difference detection and intelligent matching, comprising the following modules:

[0112] The difference detection and out-of-syllabus recognition module is used to acquire grade and region information and the text to be detected. It uses the difference detection operator to compare the lexical features of the text with the vector cosine similarity of the preset standard database, and identifies and outputs a set of out-of-syllabus words.

[0113] Contextual semantic anchor generation module: used to extract contextual features of the over-the-top word set, identify key nodes through contextual semantic operators, and generate corresponding contextual semantic anchor vectors using spatial mapping algorithms;

[0114] The current valid fingerprint modeling module is used to fuse individual feature vectors extracted from historical samples with baseline feature vectors extracted from a standard database, calculate weight coefficients through a confidence smoothing function, and generate the current valid fingerprint.

[0115] Non-original deviation risk determination module: It is used to extract the conjunction transformation matrix and semantic transition probability through syntactic dependency parsing operator and construct logical connection vector, and use manifold learning algorithm to calculate the distance between it and the current valid fingerprint to determine the non-original deviation risk;

[0116] Cognitive confirmation operator solution module: used to verify questions based on context semantic anchor vector matching. It extracts the answer accuracy and response time through the cognitive load analysis operator and calculates the deviation, and then couples the solution to obtain the cognitive confirmation operator.

[0117] Style regression adjustment and output module: It is used to generate polished text in a probability field. It uses an adaptive regression adjuster to adjust style regression based on cognitive confirmation operators and non-original deviation risk, and performs structural dimensionality reduction through syntactic weakening operators to output target text.

[0118] Specifically, the difference detection and out-of-syllabus recognition module obtains grade and region information by calling the underlying text parsing operator, performs word segmentation on the received text to be detected, maps words to the feature space using a high-dimensional vectorization operator, calculates the cosine similarity between the feature and the benchmark vector in the preset standard database through the difference detection operator, and then identifies and outputs a set of out-of-syllabus words.

[0119] The context semantic anchor generation module extracts related terms of out-of-scope words through a sliding window operator, uses a multi-head attention mechanism as a context semantic operator to identify key nodes that contribute the most to semantic orientation, and uses a spatial mapping algorithm to map the above nodes into context semantic anchor vectors with fixed coordinates in the feature space.

[0120] The effective fingerprint modeling module retrieves the user's historical sample data and dynamically adjusts the fusion weight coefficients of the individual feature vector and the baseline feature vector through a confidence smoothing function, thereby generating a current effective fingerprint that reflects the user's true writing habits.

[0121] The module for determining the risk of non-originality deviation uses syntactic dependency parsing operators to perform in-depth analysis of the text's writing logic, extracts the conjunction transition matrix and semantic transition probability, constructs a logical connection vector and uses a manifold learning algorithm to measure its distance from the current valid fingerprint, thereby determining the risk of non-originality deviation of the text.

[0122] The cognitive confirmation operator solution module retrieves matching semantically aligned test questions from a preset question bank based on contextual semantic anchor vectors. It captures the deviation between the accuracy and response time of the user's answering process through the cognitive load analysis operator, and obtains the cognitive confirmation operator that represents the individual's true cognitive ability through nonlinear coupling solution.

[0123] The style regression adjustment and output module generates initial polished text within a probability field defined by the current valid fingerprint and the set of out-of-scope words. Then, using an adaptive regression adjuster, based on the quantization value of the cognitive confirmation operator and the risk level of non-original deviation, it performs dimensionality reduction and pruning operations on sentence structures in the text that exceed the preset complexity through the syntactic weakening operator, and finally produces target text that matches the user's cognitive level.

[0124] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for automatic data verification based on difference detection and intelligent matching, characterized in that, Includes the following steps: Obtain the grade and region information associated with the user, and obtain the text to be detected uploaded by the user. By comparing it with the preset standard database, identify and output a set of words that are beyond the scope of the curriculum. Extract the contextual features of the over-the-top word set in the text to be detected, and generate the corresponding contextual semantic anchor vector; The individual feature vectors extracted from historical samples are combined with the baseline feature vectors extracted from the standard database to generate the current valid fingerprint; Calculate the distance between the logical connection features of the text to be detected and the current valid fingerprint to determine the risk of non-original deviation; The question is verified based on the contextual semantic anchor vector matching, and the cognitive confirmation operator is calculated based on the user's answer response data; The polished text is generated within a probability field defined by the current valid fingerprint and the set of out-of-scope words. The style regression adjustment is performed using the cognitive confirmation operator and the non-original deviation risk to produce the target text.

2. The automatic data verification method based on difference detection and intelligent matching according to claim 1, characterized in that, The process of obtaining the user-uploaded text to be detected includes the following steps: The system receives raw text uploaded by users through an interactive interface and uses character cleaning operators to remove non-target language characters from the raw text to obtain standardized text. The standardized text is subjected to language detection and legality verification. If the verification passes, the standardized text is input as the text to be detected into subsequent steps. The preset standard database is a syllabus vocabulary database set up based on the grade and region information.

3. The automatic data verification method based on difference detection and intelligent matching according to claim 1, characterized in that, The process of identifying and outputting the set of words beyond the scope of the syllabus includes the following steps: Lexical features are extracted from the text to be detected by a difference detection operator and compared with the vector cosine similarity of the compliant vocabulary in the preset standard database. Words not covered by the compliant thesaurus are marked as out-of-scope candidate words, and the out-of-scope candidate words are filtered and deduplicated according to the syntactic structure of the text to be detected to generate the out-of-scope word set.

4. The automatic data verification method based on difference detection and intelligent matching according to claim 1, characterized in that, The extraction of contextual features of the set of words beyond the scope of the text to be detected includes the following steps: Using a sliding window operator, context-related terms within a preset span are obtained, centered on each super-category word in the super-category word set. The context-related terms are vector-encoded using a pre-trained language model, and combined with the position index of the super-class words in the text to be detected, contextual features representing the semantic environment of the super-class words are calculated.

5. The automatic data verification method based on difference detection and intelligent matching according to claim 1, characterized in that, The steps for generating the corresponding context semantic anchor vector are as follows: By assigning weights to the contextual features using contextual semantic operators, the key nodes that contribute the most to the semantic orientation of the super-category word set are identified. The key nodes are converted into contextual semantic anchor vectors with specific coordinate dimensions in the feature space using a spatial mapping algorithm.

6. The automatic data verification method based on difference detection and intelligent matching according to claim 1, characterized in that, The process of generating a currently valid fingerprint includes the following steps: The weighting coefficients corresponding to the sample throughput of the historical samples are calculated using the confidence smoothing function; Based on the weighting coefficients, the individual feature vector and the baseline feature vector are linearly or non-linearly weighted and fused to generate a current valid fingerprint representing the corresponding user's capability characteristics.

7. The automatic data verification method based on difference detection and intelligent matching according to claim 1, characterized in that, The calculation of the distance between the logical connection features of the text to be detected and the current valid fingerprint includes the following steps: The conjunction transition matrix and semantic transition probability in the text to be detected are extracted by the syntactic dependency parsing operator. Construct a logical connection vector and use a manifold learning algorithm to calculate the manifold distance between the logical connection vector and the current valid fingerprint in the feature space.

8. The automatic data verification method based on difference detection and intelligent matching according to claim 1, characterized in that, The process of obtaining the cognitive confirmation operator based on user response data includes the following steps: The cognitive load analysis operator is used to extract the user's correct answer rate and response time for the verification questions, and the deviation between the response time and the preset standard time is calculated. The accuracy rate of answering questions and the deviation rate are coupled and calculated to generate a cognitive confirmation operator that represents the user's certainty in mastering advanced vocabulary.

9. The automatic data verification method based on difference detection and intelligent matching according to claim 1, characterized in that, The style regression adjustment using the cognitive confirmation operator includes the following steps: The quantized value of the cognitive confirmation operator is obtained through an adaptive backoff regulator; If the value is lower than a preset threshold or the risk of non-original deviation exceeds a preset threshold, the smoothness of logical connection of the polished text is reduced according to the inverse weight of the value or the risk. By using syntactic weakening operators to reduce the structural dimensionality of sentence structures with a preset complexity or higher in the polished text, the final target text is reduced in terms of logical complexity and pragmatic style to the cognitive range represented by the current effective fingerprint.

10. An automatic data verification system based on difference detection and intelligent matching, characterized in that, The automatic data verification method based on difference detection and intelligent matching as described in any one of claims 1-9 includes the following modules: The difference detection and out-of-syllabus recognition module is used to acquire grade and region information and the text to be detected. It uses the difference detection operator to compare the lexical features of the text with the vector cosine similarity of the preset standard database, and identifies and outputs a set of out-of-syllabus words. Contextual semantic anchor generation module: used to extract contextual features of the over-the-top word set, identify key nodes through contextual semantic operators, and generate corresponding contextual semantic anchor vectors using spatial mapping algorithms; The current valid fingerprint modeling module is used to fuse individual feature vectors extracted from historical samples with baseline feature vectors extracted from a standard database, calculate weight coefficients through a confidence smoothing function, and generate the current valid fingerprint. Non-original deviation risk determination module: It is used to extract the conjunction transformation matrix and semantic transition probability through syntactic dependency parsing operator and construct logical connection vector, and use manifold learning algorithm to calculate the distance between it and the current valid fingerprint to determine the non-original deviation risk; Cognitive confirmation operator solution module: used to verify questions based on context semantic anchor vector matching. It extracts the answer accuracy and response time through the cognitive load analysis operator and calculates the deviation, and then couples the solution to obtain the cognitive confirmation operator. Style regression adjustment and output module: It is used to generate polished text in a probability field. It uses an adaptive regression adjuster to adjust style regression based on cognitive confirmation operators and non-original deviation risk, and performs structural dimensionality reduction through syntactic weakening operators to output target text.

Images