67 results about "Semantic equivalence" patented technology

The invention provides a source database storage process migration method and device, electronic equipment and a storage medium, and the method comprises the steps: carrying out the grammar analysis of an original storage process in a source database, and generating a corresponding abstract syntax tree; based on a preset rewriting rule set, performing semantic equivalent transformation on the flow control syntax structure in the abstract syntax tree to generate a transformed abstract syntax tree; and generating a target script in a target execution environment from the converted abstract syntax tree, executing the target script in a script running engine, dynamically generating a standard SQL statement corresponding to the business logic defined by the original storage process, and submitting the standard SQL statement to one or more target data platforms for execution. And the business logic of the original storage process is reproduced. By introducing the target script as the intermediate execution layer, the flexibility of the migration scheme is greatly improved, so that the migration scheme is suitable for various big data platforms without the native storage process capability.

The disclosed embodiments describe a method, system, and computer-readable medium for generating a training dataset for training a model in the field of natural language processing involving receiving a set of input samples and performing a rephrasing operation to produce new versions of the set of input samples, where the new versions preserve semantic equivalence as the set of input samples but have different phrasing. A dataset of generated versions of the input samples is generated using a generative Language Learning Model (LLM), all entity references present in the generated versions of the input samples are labeled, and the generated versions of the input samples and their corresponding labeled versions to form an expanded labeled dataset are aggregated.

The invention discloses a WebShell malicious behavior fine grit method based on an abstract syntax tree and de-obfuscation, which comprises the following steps: inputting a WebShell suspicious script file into a computer system, and quickly identifying and normalizing the suspicious script through file format screening, feature matching and file standardization to obtain a normalized script; executing a de-obfuscation operation on the standardized script based on a large language model to obtain a simplified semantic equivalence code; analyzing the code to generate an abstract syntax tree, extracting key features through a multi-granularity feature extraction mechanism, and capturing multi-dimensional context dependency; and extracting multi-dimensional feature vectorization, performing feature fusion, inputting a deep neural network combined with ResNet, BiLSTM and Transform, and outputting a WebShell malicious behavior detection result and a malicious behavior fine-grained classification result. The WebShell attack detection method can solve the problems of low detection accuracy and high false alarm and missing report rate of an existing detection method when facing highly hidden, deformed and confused WebShell attacks.

The invention relates to the technical field of body intelligence, in particular to a body intelligence attack detection method based on a constraint function, mainly solves the technical problems of insufficient granularity, poor real-time performance and insufficient robustness of the existing attack detection method, and comprises the following steps: S1, selecting key points; s2, generating a constraint function; s3, analyzing the images where the key points are located by adopting a credible multi-modal model, and generating task fingerprints; s4, comparing the constraint function with the task fingerprints, and calculating an output difference; semantic equivalence measurement is introduced; a tolerance value is introduced, if the output difference exceeds a threshold value, it is judged that the constraint function is attacked, and the system stops task execution. According to the method, the detection accuracy exceeding 92% can be realized in various attack scenes; the problems of false report and missing report caused by insufficient rule coverage or large language model output diversity can be avoided, and the robustness is high; and the execution safety and the real-time performance of the body-equipped intelligent agent in a complex and open environment can be obviously improved.

This application discloses a method, system, and device for detecting malicious code variants using control flow graph semantic matching, belonging to the field of computer network security technology. It includes: disassembling the target executable file, constructing a control flow graph, and extracting basic block semantic features to generate semantic vectors; obtaining semantic equivalence classes based on semantic vector clustering, identifying main anchor points, and constructing a main anchor point skeleton graph; performing semantic compression and noise processing with the anchor point skeleton as constraints, and obtaining a semantic core graph through topological verification; hierarchically normalizing the semantic core graph and the main anchor point skeleton graph to generate a composite hash signature; and matching the composite hash signature with a malicious code family signature library to complete variant determination. The above scheme, through the main anchor point constraint compression and normalization process, resists structural obfuscation disturbances, improves the consistency of homologous variant identification, and has advantages such as strong robustness, high efficiency, and good interpretability. It is suitable for detecting malicious code variants using control flow graph semantic matching in complex obfuscated environments.

This invention relates to the field of data management technology, specifically to a method, apparatus, computer equipment, and storage medium for multi-source data consistency management. The method includes: acquiring data records from at least two source systems to generate change events; converting the change events into evidence records containing original values ​​and contextual information based on a semantic rule registration library, and generating a standardized value view; performing entity parsing based on the standardized value view to obtain a set of candidate entity matching relationships, and outputting a credible candidate set through semantic equivalence transformation stability testing and evidence conflict measurement; performing constraint consistency inference based on the credible candidate set and the semantic rule registration library to generate a consistency state snapshot and a correction action plan; generating a synchronous execution rule set based on the correction action plan and performing a pre-rehearsal acceptance test; executing correction and re-collecting verification when successful, and rolling back when unsuccessful. This invention improves the traceability of cross-source consistency determination and the verifiability of correction execution, and reduces the risk of erroneous corrections.

The application discloses a kind of semantic equivalent text generation methods based on familiar word constraint. Among them, the method includes: obtaining user familiar word set, and based on the familiar word set, construct structured limited expression space, for limiting the range of vocabulary and expression structure range that can be used in text generation process;Target text is carried out semantic structure analysis, and the target semantic structure representation is generated;In the structured limited expression space, semantic matching solution is executed, and text generation problem is converted into limited semantic solution problem, at least one expression path that satisfies the target semantic structure representation is constructed under the condition of the limited expression space;Expression candidate set is generated based on the expression path;The semantic consistency determination is carried out to the expression candidate set, and the semantic equivalent expression variant set that satisfies consistency condition is output.The application solves the technical problem that existing text generation technology is difficult to structure solution for target semantics within limited vocabulary range.

The invention discloses a large model reasoning stability enhancement method and system in a text scene, and relates to the technical field of knowledge enhancement deep learning, and the method comprises the steps: building a structured text knowledge graph based on a text knowledge base, and generating a dynamic text knowledge embedding matrix through a graph attention network; inserting a text knowledge gating cross attention module into a decoder selection layer of the pre-trained large model, taking the text knowledge gating cross attention module as an external knowledge source, obtaining a knowledge enhanced hidden state after gating fusion, and constructing a transformation model; semantic equivalent perturbation is carried out on an input text to obtain a perturbation sample, the perturbation sample is input into the transformation model in parallel to obtain extraction probability distribution, and divergence and gradient direction consistency loss between two distributions are calculated; and combining cross entropy loss and gradient direction consistency loss to train and transform the model, and updating parameters to convergence to obtain a final large model. According to the method, the fact consistency of output can be improved, common optimization of knowledge guidance and stability constraint is realized, and the result is accurate and reliable.

This invention discloses a detection method based on a large language model, belonging to the field of natural language processing technology. The method includes: collecting and preprocessing the original detection text; generating detection prompts by constructing dynamic contextual anchors based on the preprocessed data through semantic scoring and redundancy suppression; obtaining the main original detection response through the large language model based on the detection prompts, constructing semantically equivalent perturbation text, and evaluating the stability of the large model's detection response; parsing the response conclusions and calculating the perturbation consistency confidence score; making a detection judgment based on the perturbation consistency confidence score and outputting structured detection results; and performing self-calibration of the confidence score threshold using the structured detection results. This method not only significantly improves the accuracy, interpretability, and robustness of detection but also effectively enhances resistance to user rewriting avoidance behavior and reduces the risk of detection errors caused by missing context or superficial wording changes.

This invention relates to the field of natural language processing technology, and more particularly to a method and system for language transfer recognition based on cultural semantic embedding. The method includes the following steps: acquiring source language data and target language data; performing semantic isomorphic centering on the source and target language data to obtain semantically equivalent data; decoupling pragmatic features based on the semantically equivalent data to obtain pragmatic feature data; performing contrastive cultural latent space embedding based on the pragmatic feature data to obtain bilateral cultural representation data; performing cultural modality affine transformation based on the bilateral cultural representation data to obtain cultural difference data; performing implicit distortion pattern discrimination based on the cultural difference data to obtain cultural semantic distortion data; and generating transfer results based on the cultural semantic distortion data to obtain language transfer result data. This invention decouples pragmatic features under semantic equivalence constraints and achieves language transfer recognition through cultural semantic embedding and modality affine modeling.

The application discloses a kind of based on semantic context's maintenance information aggregation search method and system, it is related to natural language processing technical field;After text analysis to search request, based on the maintenance semantic context extraction of preset static maintenance semantic label library obtains dynamic context;After the structured organization construction of dynamic context and search request, multiple semantic equivalent rewriting is carried out to obtain multiple rewritten search requests;Based on the scheme maintenance result search of preset multi-source maintenance knowledge base to all rewritten search requests obtains search result set;Based on dynamic context, search result set is sorted, and the first preset search result is obtained as final output.Analysis user natural language request, then rely on static maintenance semantic label library to complete dynamic context, then by structured fusion and multi semantic rewriting search, aggregate multi-source knowledge without omission, to quickly obtain the accurate scheme of adaptation, improve search accuracy while also greatly improve search efficiency.

The application provides a large model NL2SQL evaluation method and device based on database exploration, and relates to the technical field of data processing. The method constructs the semantic correctness evaluation process of NL2SQL as a multi-step decision process driven by a large model agent. The large model evaluation agent module generates a probing query by calling a read-only database execution tool module, iteratively updates the context according to the dynamically returned execution results, and outputs a semantic equivalence judgment. The present scheme changes the traditional "static text matching" evaluation paradigm to "dynamic execution verification". Through the closed-loop feedback mechanism of "proposing a hypothesis - executing verification - updating evidence", the evaluation bias caused by a single reference SQL is effectively solved, complex cases such as "accidental consistency of execution results but semantic error" or "inconsistency of execution results but semantic correctness" can be identified, and the execution accuracy of the evaluation and the robustness of the semantic discrimination are significantly improved.

The invention belongs to the technical field of network security, and particularly relates to a Web application firewall vulnerability detection method based on load variation injection. Comprising the following steps: constructing a unified attack grammar rule base covering multiple attack types; analyzing the original attack load into an abstract syntax tree; carrying out semantic reservation load variation based on the syntax tree to generate a semantic equivalent variation load; performing protection rule detection on the Web application firewall, analyzing a detection result and performing pruning operation on the syntax tree; a Monte Carlo tree search algorithm is utilized, a rule detection mechanism is combined, and a variation path with a high success rate is dynamically evaluated and preferentially tested so as to efficiently generate an antagonistic load capable of bypassing WAF detection; according to the method, the problems of attack failure, low black box test efficiency and insufficient coverage of a single attack type caused by semantic destruction in the prior art are solved, and the WAF rule base vulnerability can be systematically evaluated.

The invention discloses an intelligent question and answer method and system based on double-stage retrieval and generation. The method comprises: receiving a user query; vectorizing the questions through a mixed retrieval model, firstly retrieving in a standard knowledge base, and directly returning corresponding standard answers if semantic equivalence standard questions are found; otherwise, retrieving again in the policy original text knowledge base, and generating a final answer through a retrieval enhancement generation technology based on a retrieval result. Wherein the hybrid retrieval model is formed by carrying out weight fusion on the results of the original Embedding and the fine adjustment Embedding. According to the system, high-frequency standard questions and low-frequency complex questions are distinguished through a double-stage mechanism, the response efficiency and answer consistency are guaranteed, and meanwhile complex consultation needing policy original text reference can be processed. The hybrid retrieval model uniformly supports two-stage retrieval, takes into account general semantic understanding and domain optimization, and improves the retrieval quality and the system stability.

The invention provides an intelligent heterogeneous CAD platform code migration method and system based on a knowledge graph, and relates to the technical field of knowledge graphs, the method comprises the following steps: constructing semantic representation by extracting implicit constraints called by a source platform code API, identifying paradigm differences by means of the knowledge graph and applying a conversion rule, and carrying out conversion processing on geometric kernel, parameterization and state management differences, and carrying out semantic equivalence verification. After the verification is passed, constructing a migration sub-graph and extracting a high-frequency mode to form a reusable rule; and positioning the deviation source and updating the rule when verification fails. According to the method, the automatic migration of the codes among the heterogeneous CAD platforms is realized, and the migration efficiency and accuracy are improved.

The invention discloses a marine economy sustainable development capability assessment method based on big data processing, particularly relates to the technical field of big data processing, and is used for solving the problems that an intermediate calculation result cannot be reused, calculation is repeated and response is delayed due to structure mismatch of batch calculation and instant query in existing marine economy assessment. Intermediate calculation results and data features and semantic attributes of the intermediate calculation results are extracted by analyzing historical batch calculation jobs, active reasoning is performed, a relation graph representing semantic association of the intermediate calculation results is generated, and an intermediate calculation result catalogue with a uniform interface is constructed according to the relation graph. Performing semantic equivalence reconstruction on the instant query target calculation diagram based on the historical calculation motif frequency spectrum to generate an optimization calculation logic diagram, mapping the optimization diagram and a catalogue, identifying a calculation sub-diagram capable of directly providing a result, finally executing the optimization diagram, obtaining the result by calling a catalogue interface, and completing residual calculation to respond to query. And intelligent identification and efficient reuse of historical intermediate calculation results are realized.

The invention provides an artificial intelligence-based material main data intelligent cleaning method, which comprises the following steps: obtaining a material description text, extracting text features through a multi-modal feature extraction strategy, and disassembling material description according to a classification template, the material description comprising a technical name, a specification parameter, a measurement unit and material data; material description is obtained, equivalent description conversion is carried out through a corresponding synonym database, missing attributes are pre-judged based on the industry knowledge graph, and material information is generated; material information of the multiple groups of material description texts is obtained, various descriptions in the material information are compared, if the material descriptions are in one-to-one correspondence, the material information is combined, a counting strategy is preset, and the number of the combined material information is obtained; by obtaining original material description and AI material description subjected to AI processing, the difference between the original material description and the AI material description is analyzed, a semantic equivalence rule is generated, and a synonym database is updated.

The invention belongs to the technical field of software security, and particularly relates to a malicious software automatic analysis-oriented anti-obfuscation binary function name recovery method, which comprises the following steps of: firstly, constructing a triple data set comprising obfuscation assembly codes, clear assembly codes corresponding to the obfuscation assembly codes and real function names; then utilizing the data set to finely adjust a large-scale language model, so that the large-scale language model has the translation capability of converting obfuscated codes into semantic equivalent clear codes; secondly, mapping the clear assembly code into a semantic feature vector with a fixed dimension by using an anti-confusion graph semantic encoder; and finally, constructing a function name decoder, and carrying out joint training and optimization on coding and decoding parts by minimizing the loss between a prediction function name and a real function name.

This invention relates to the field of natural language processing technology, and more particularly to a method and system for language transfer recognition based on cultural semantic embedding. The method includes the following steps: acquiring source language data and target language data; performing semantic isomorphic centering on the source and target language data to obtain semantically equivalent data; decoupling pragmatic features based on the semantically equivalent data to obtain pragmatic feature data; performing contrastive cultural latent space embedding based on the pragmatic feature data to obtain bilateral cultural representation data; performing cultural modality affine transformation based on the bilateral cultural representation data to obtain cultural difference data; performing implicit distortion pattern discrimination based on the cultural difference data to obtain cultural semantic distortion data; and generating transfer results based on the cultural semantic distortion data to obtain language transfer result data. This invention decouples pragmatic features under semantic equivalence constraints and achieves language transfer recognition through cultural semantic embedding and modality affine modeling.

The invention discloses a member leakage risk assessment method and system for a retrieval enhancement generation system based on a small amount of queries. Under the black box condition, according to the target member document, generating a plurality of queries with equivalent semantics and diverse syntaxes by the auxiliary large language model, submitting the queries to a target retrieval enhancement generation system, and collecting responses; the response is encoded as a sentence vector, an average pairwise similarity is calculated and a rejection response is identified to apply a penalty, and a final risk assessment score is synthesized and compared with a threshold to assess the likelihood of the target document as a knowledge base member. The method is implemented under the complete black box condition, does not depend on the internal architecture or parameter information of the system, is low in query overhead, is easy to calibrate, and is suitable for compliance auditing, red team evaluation and security detection before online, so as to quantify the member privacy leakage risk and guide a protection strategy.

Embodiments described herein are directed to generating and returning materialized views for queries (or subexpressions thereof) having a particular relationship with each other. For instance, machine learning-based techniques may be utilized to identify query subexpressions that have at least one of a semantically equivalent relationship or a containment relationship with each other. Responsive to identifying such relationship(s), a materialized view may be generated for the identified subexpressions. When a query is subsequently received, machine learning-based techniques may be utilized to determine whether a subexpression of the query possesses at least one of a semantically equivalent relationship or a containment relationship with another subexpression for which a materialized view has been generated. Responsive to determining that such a subexpression of the query possesses one or more of such relationships, the materialized view generated for the other subexpression is returned.

The application relates to the technical field of code intelligent analysis, and provides a source code vulnerability intelligent detection method and system, which comprises the following steps: obtaining function-level source code text to be detected, generating a word unit sequence through a word segmenter, extracting a hidden state after length alignment; extracting three-granularity features through a multi-granularity feature extraction module, obtaining fusion features through an adaptive gate fusioner; the adaptive gate fusioner scores the three-granularity features through a gate network, then performs weighted fusion on the three-granularity features to obtain fusion representation, and obtains a vulnerability probability through a classifier. In the training stage, a double-view enhanced input of semantic equivalence is introduced, a closed-loop training mechanism is constructed by jointly optimizing cross-entropy loss, supervised contrast loss and consistency distillation loss, and a weighted sampling strategy based on the reciprocal of the category frequency is adopted. The application can effectively capture local and cross-scale vulnerability patterns, and improve the recognition ability and detection stability of vulnerabilities triggered by local sentences.

Examples detect equivalent subexpressions within a computational workload. Examples include converting a query plan tree associated with a first subexpression into a matrix. The first subexpression is a portion of a database query from the computational workload. Each node in the query plan tree is represented as a row of the matrix. The matrix is converted into a first vector. The first subexpression is determined to be equivalent to a second subexpression by comparing the first vector to a second vector associated with the second subexpression. The comparison includes computing a distance between the first and second vectors that is lower than a distance threshold. The computational workload is modified, based on the determining, to perform the first subexpression and exclude performance of the second subexpression as duplicative.

The application discloses a kind of intelligent labeling method, equipment and medium of text data, belong to electric digital data processing technical field, comprising: obtaining the label ontology of source field and target field;Based on the concept granularity, semantic boundary and axiom form between the two, construct the semantic alignment graph of cross ontology;According to the formal semantic equivalence determination of the logic rule of the graph to source field, generate the translated logic rule suitable for target field;After multi-label prediction is carried out to target text, the consistency check is carried out to candidate label combination using translation rule, and the final labeling result is output.The application solves the logic rule invalidation problem caused by label system isomerism and semantic boundary dynamic change in cross-domain scene by constructing the migration framework based on formal ontology semantic alignment and hierarchical translation verification, realizes the semantic fidelity and structure consistent translation of logic rule between heterogeneous knowledge system, effectively avoids semantic drift and conflict misjudgment.

The invention discloses an NL2SQL fine-tuning optimization method based on deep learning. The method comprises the following steps: firstly, acquiring a natural language question, generating a target SQL in combination with similar examples, labeling elements, recording complexity, and strengthening model learning through data enhancement; then, an integrity index is generated through analysis, an adjustment object is determined according to a threshold value and semantic equivalence, meanwhile, a correct rate index is analyzed, and a corresponding signal is generated; and finally, collecting database information, and optimizing the SQL execution efficiency based on the execution plan. According to the NL2SQL fine-tuning optimization method based on deep learning provided by the invention, the semantic equivalent NL2SQL is classified and processed by adopting the hierarchical adjustment and dynamic adaptation strategy and the equivalent group strategy, so that optimization strategy multiplexing is realized, repeated calculation is greatly reduced, and the optimization efficiency in a large-scale scene is improved. The generalization ability of the model is improved through data enhancement, and the generation quality and execution efficiency of the NL2SQL are comprehensively guaranteed.

The invention discloses a semantic confusion-based retrieval method, system and device and a medium, and the method comprises the steps: before and after endowing retrieval information, sampling a given question based on a preset large language model, and obtaining a corresponding answer result and a generation probability; all the answer results expressing semantic equivalence are merged into the same semantic class set; calculating the sum of the generation probabilities of all the answer results in the semantic class set, and taking the sum as the semantic confusion of the large language model for the given question; measuring the contribution degree of the retrieval information by analyzing the gain between the determined semantic perplexity before and after endowing the retrieval information with the retrieval information; measuring the contribution degree of the retrieval information by analyzing the gain between the determined semantic perplexity before and after endowing the retrieval information with the retrieval information; and performing optimization training on the large language model according to an enhanced sample data set constructed by the contribution degree to improve the model output quality.

The invention provides a data processing method of a database and a related product. The data processing method comprises the following steps: acquiring a to-be-processed data set, and determining a grouping column and a ranking sequence; grouping tuples of the data set according to the grouping key values of the grouping columns by using the hash table; sorting the tuples in each group by using a heap sorting structure, and taking a preset number of tuples at the sorting head in each group as target tuples; and forming a sorting array by the target tuples in each group to obtain a result set. According to the data processing method, on the premise of ensuring semantic equivalence, the sorting calculation amount and memory occupation in window function query are greatly reduced, the query execution performance and efficiency are improved, and the method is particularly suitable for a high-frequency Top-N scene.