19 results about "Semantic pattern" patented technology

The invention discloses an aspect-level sentiment analysis optimization method based on multivariate external knowledge fusion, which relates to the technical field of sentiment analysis optimization, and comprises the following steps: constructing a multivariate external knowledge source comprising a Chinese sentiment dictionary, a domain knowledge graph and a user comment prior mode library; the sentiment module is used for providing vocabulary-level sentiment polarity, entity attribute relations and high-frequency evaluation semantic modes; semantic coding is performed on the input text and the specified aspect words to generate context semantic representation, and global semantic features and local position features are extracted in combination with aspect word position information; based on a semantic coding result, converting the multivariate external knowledge sources into structured knowledge representations, and dynamically adjusting contribution weights of various types of knowledge through a gating fusion mechanism to generate fused knowledge representations; performing dependency syntactic analysis on the input text, constructing an original syntactic structure, and calculating the correlation strength of each grammatical component and aspect words in combination with context semantic representation; and pruning the original syntactic structure according to the correlation intensity.

This invention discloses a computer network security protection method and system based on data analysis, relating to the field of computer security technology. Targeting IT / OT converged factory networks, this invention aligns traffic, DNS, authentication, processes, topology, and industrial instructions into session chains, generating semantic patterns and semantic constraints. It enables baseline establishment based on semantic patterns, segmented output of evidence chains and risk scores, and the linkage of semantic enhancement parameters with acquisition frequency, rate limits, and micro-isolation range. This allows the control end to handle only the abnormal segments with minimal intervention, reducing false alarms and production disruption risks, and enabling closed-loop update strategies to be transmitted back.

The present application relates to the technical field of information security, and aims at the problems that password guessing based on probabilistic context-free grammar is difficult to identify multi-semantic patterns and the semantic guidance strength is uncontrollable, and proposes a password guessing method based on multi-semantic fusion of probabilistic context-free grammar: enumerating sub-strings in the training password, identifying semantic segments according to simple, date, vocabulary and name patterns and prioritizing disambiguation, dynamically planning segmentation according to the principle of maximum semantic coverage and least semantic segments, setting semantic enhancement parameters β for each semantic pattern during training to weight and normalize the count, and outputting the candidate password dictionary according to the probability priority queue during generation, which is suitable for efficient password guessing in offline password audit, password policy evaluation and penetration testing.

The application provides a power transmission line naming recognition method based on a bidirectional enhanced nonlinear pulse nerve, and the core innovation is that a BiENSNP module is constructed to deeply simulate a dynamic information processing mechanism of a biological nerve system, the model is endowed with powerful basic representation capability, and is especially good at modeling complex nonlinear semantic patterns contained in text. The fusion architecture significantly improves the robustness and accuracy of the model in identifying entities in the power transmission line construction field (especially generative text), thereby laying a solid and good scalable technical foundation for constructing high-quality power transmission line construction knowledge base and other key application scenarios.

The application provides a large model KV cache multi-dimensional compression method and system for long text tasks, which comprises the following steps: sampling input samples capable of covering the context structure and semantic mode of a large language model from long text task related corpus to construct a calibration dataset; loading the pre-trained large language model into an inference framework, performing forward inference using the calibration dataset, extracting each layer of KV cache generated in the forward inference process, and storing it by level; performing singular value decomposition on the KV cache of each layer and calculating the energy proportion of the first r singular values to determine the low rank degree of the KV cache of the layer; performing joint compression in the rank dimension and the quantization dimension according to the determined low rank degree; and using the KV cache of all layers after joint compression for inference deployment. The application realizes efficient compression and precision maintenance of the large language model KV cache in long text tasks.

The embodiment of the invention discloses a data classification method based on a large model, and the method comprises the steps: obtaining unlabeled and labeled narrative text data, recognizing unknown category data which does not belong to the existing labels through clustering, generating the semantic description of the unknown category based on a high-confidence sample by using a large language model, and carrying out the classification of the unknown category. And calculating the distinction degree between the description and the existing category, when the distinction degree is insufficient, carrying out semantic calibration by taking the annotated data as a reference, and finally, completing classification of all data according to the optimized semantic description. By introducing the semantic understanding ability of a large language model, a traditional clustering result based on numerical features is converted into an adjustable semantic mode, the problem that the clustering result in complex narrative text classification is inconsistent with an artificial standard is effectively solved, and meanwhile, the classification efficiency is improved. Semantic calibration also improves the accuracy and reliability of new category discovery in a category imbalance scene.

The invention discloses a one-dimensional discrete sequence anomaly detection and repair method and system, and belongs to the technical field of data processing, and the method comprises the steps: carrying out the preprocessing of a one-dimensional discrete sequence, and generating an input sequence; performing forward propagation on the input sequence by using a mask language model to obtain category probability distribution of each position; calculating an anomaly score based on the probability distribution; comparing the abnormal score with a preset threshold value, and generating an abnormal position mask; and repairing the abnormal position, and using the maximum probability category in the probability distribution as a repairing value. According to the method, anomaly detection and repair are unified into a mask language model task, so that integrated end-to-end processing of detection and repair is realized, a threshold value does not need to be manually set, a repair result conforms to a global semantic mode, and the accuracy and the automation degree of anomaly processing are remarkably improved.

The invention belongs to the technical field of communication, and particularly relates to an integrated system of semantics on short waves based on low-speed waveforms and a use method. According to the method, the semantic communication technology and short-wave communication are deeply fused, and the strong robustness of semantic information to channel noise and distortion and the technical capabilities of semantic compression, intelligent coding and semantic reconstruction are utilized, so that transmission logic innovation of directly transmitting information semantic content is realized; through actual verification of a prototype, on the premise that the audio quality is not reduced, the requirement of a short-wave radio station in a semantic mode for the minimum signal-to-noise ratio of communication is reduced by more than 2dB (inclusive), the anti-interference capability of short-wave communication is greatly improved, the communication passability in a complex environment is effectively improved, and the communication efficiency is improved. And the performance weakness of the traditional short-wave communication in a high-interference scene is directly compensated. According to the method, the semantic communication technology is applied to the short-wave radio station for the first time, preliminary application verification is completed, and a breakthrough of semantic communication from theory to short-wave actual equipment is successfully achieved.

The invention discloses a semantic analysis method for a dynamic compensation neural network. The semantic analysis method comprises the steps of 1, input preprocessing and basic feature extraction, 2, initial semantic analysis and error evaluation, 3, execution of error compensation and 4, semantic fusion and output after compensation. According to complexity and domain characteristics of input semantics, compensation nodes can be adaptively activated, connection weights can be adjusted, multi-scene semantic requirements can be flexibly adapted, a real-time error feedback mechanism effectively inhibits underlying feature extraction deviation and analysis errors of a rare semantic mode, robustness is enhanced, a lightweight basic network is combined with an on-demand compensation strategy, and multi-scene semantic requirements can be met. The reasoning efficiency is optimized, the response requirement of a real-time interaction scene is met, high-precision, high-efficiency and high-adaptability semantic analysis capacity is finally achieved, and the method can be widely applied to natural language processing scenes such as intelligent question answering and dialogue systems.

The application relates to the technical field of natural language processing, in particular to an intelligent foreign trade data classification system based on natural language processing, which comprises a semantic pattern recognition module, a classification task execution module, a classification effect monitoring module, a cross-domain data migration module and a rule optimization feedback module.In the application, semantic analysis is carried out based on language features and context association rules, the semantic structure of foreign trade data can be effectively recognized and classified, the dependence on manual rules in the classification process is reduced, the accuracy and automation degree of classification are improved, the task allocation path can be optimized according to actual requirements, abnormal data can be identified in time and classification rules can be adjusted by monitoring the classification effect, the overall classification quality is improved, the cross-domain data migration and rule optimization feedback mechanism ensure that the classification rules can be adapted in different fields, the flexibility and expansibility of the system are enhanced, and the overall efficiency and precision of foreign trade data processing are effectively improved.

The application discloses a personalized recommendation method and system based on large language model and field model cooperation, and relates to the technical field of information recommendation. In the target recommendation scene, the method processes unstructured data through a large language model to obtain semantic mode features, and simultaneously analyzes user quantifiable operation records by using a specified field model to output behavior mode features, thereby realizing the synchronous extraction of unstructured semantic information and structured behavior information. Two types of features are subjected to bidirectional information supplement and knowledge transfer to construct a cooperatively optimized feature mapping set, thereby realizing the unified conversion and synchronous scheduling of cross-modal features. In combination with real-time interaction information reflecting the current intention of a user and the change of scene demand, an initial recommendation list is generated through double-model cooperative reasoning, and after dynamic sorting, it is determined whether to output a personalized recommendation result, thereby realizing the accurate adaptation of real-time demand of a cross-scene user and personalized recommendation, improving the timeliness of recommendation, and further effectively improving the rapid adaptability of personalized recommendation.

The application discloses a bilingual semantic neural network compiling system and method, relates to the technical field of deep learning, and can receive an ONNX model derived from multiple deep learning frameworks, automatically generate an MLIRScript representation of a Python static subset, and provide executable debugging and MLIR conversion functions in an execution semantic mode and a compiling semantic mode respectively; in the execution semantic mode, intermediate tensor values, shapes and numerical distributions can be observed in real time, and model debugging and verification can be realized; in the compiling semantic mode, a standard MLIR high-level dialect representation is generated through static analysis, the operator topology, tensor types, shapes and parameters are kept consistent with the original ONNX model, and a basis is provided for subsequent multi-hardware platform optimization. The application realizes a neural network compiling path with unified input of multiple front ends and efficient deployment of multiple back ends, solves problems, such as difficult debugging and verification, framework binding limitation and model semantic loss, in the prior art, and improves the debuggability and portability of deep learning models across frameworks and hardware.

The present disclosure proposes an active defense method for image protection, comprising: receiving a real image; adding hidden information to the received real image to generate a first image; generating a semantic pattern in a specified region of the first image based on the hidden information to generate a second image; and adjusting the hidden information to make the semantic pattern similar to a predefined semantic pattern and make the second image similar to the real image. A corresponding active defense device for image protection is also disclosed.

The invention provides a Token budget reasoning scheduling and caching integrated method and device, and belongs to the field of artificial intelligence and distributed computation.The method comprises the steps that a request is structured into a problem portrait, and an accurate key, a semantic key and an intention key are generated based on the portrait and judged according to a scene threshold value; entity mapping, timeliness and logic checking are carried out on the requests falling into the adaptation interval, and result-level adaptation multiplexing is achieved through a semantic association graph; outputting a Token budget according to a historical log, calculating a quota, and carrying out key information extraction and hierarchical compression under the constraint of the quota; the budget, the compression ratio and the real-time load are incorporated into a multi-target scheduling function for instance allocation; and updating a result by adopting LRU-K and dynamic TTL, and carrying out asynchronous preheating and index registration on a new semantic mode. Under the condition that the model structure is not changed, collaborative optimization of semantic multiplexing, response time delay and Token cost is achieved.

The invention discloses a hot-line work safety wearing time sequence semantic recognition and alarm method and system based on AR glasses, and relates to the technical field of hot-line work safety detection.The method comprises the steps that work scene video streams are continuously collected through an AR glasses camera, and a specified time sequence video clip is extracted through a sliding window; detecting the operating personnel and the protective equipment frame by frame by using the first neural network model, and outputting the equipment category and existence confidence; inputting the time sequence fragment and the detection result into a second neural network model based on a Transform architecture, analyzing continuous frame actions and equipment state changes, and identifying a behavior semantic mode; and outputting a risk level according to the semantic mode and triggering a corresponding early warning signal. According to the invention, the problem of false alarm and missing alarm of single-frame detection is solved, real-time performance and accuracy of protection monitoring are realized, and live working safety is guaranteed.

The application relates to the technical field of safety detection, and discloses a high-risk finished oil depot anti-static comprehensive detection system based on multi-sensor fusion, which comprises the following steps: multi-modal static electricity data acquisition and preprocessing, construction of a charge semantic-topology dual-communication theory framework, application of a semantic-labeled electric field topology representation algorithm, application of a topology-guided semantic analysis technology, construction of a space-time semantic-topology collaborative reasoning framework, realization of a double-level anomaly detection system, and construction of a semantic-topology knowledge accumulation mechanism. The application realizes the unification of electric field topology characteristics and charge behavior semantic modes, solves the information island problem, improves a full-dimension risk perception detection rate by 40%, reduces a false alarm rate by 85%, prolongs a risk early warning advance period to more than 10 minutes, reduces system complexity by 40%, improves calculation efficiency by 65%, and simultaneously supports experience precipitation and knowledge migration.

The invention discloses a model updating method and a related product. The method comprises the steps of obtaining to-be-processed training data; the to-be-processed training data comprises unlabeled data and labeled data; performing clustering processing on unlabeled data in the to-be-processed training data to obtain multiple groups of unlabeled data clusters; generating a label for each group of label-free data clusters in the to-be-processed training data to obtain the training data after the first processing; carrying out expansion processing on data with labels in the training data processed for the first time to obtain the training data after expansion processing; and carrying out updating training on the model by utilizing the training data after expanding writing. According to the method, the dependence of traditional supervised learning on large-scale manual annotation is effectively relieved, and the annotation cost and the iteration period are remarkably reduced; meanwhile, by introducing a long-tail semantic mode in the label-free data, the generalization ability of the model for unseen expressions is enhanced, the recognition effect on low-frequency intentions is improved, the efficiency of generating new data in the model updating process is improved, and the cost is reduced.

The invention relates to the technical field of information security, and provides a password guessing method of a probability context-independent grammar based on multi-semantic fusion aiming at the problems that the password guessing of the probability context-independent grammar is difficult to specifically recognize a multi-semantic mode and the semantic guidance intensity is uncontrollable. Semantic segments are recognized according to simple, date, vocabulary and name modes and subjected to priority disambiguation, segmentation is dynamically planned according to the segmentation principle of maximum semantic coverage and minimum semantic segments, a semantic enhancement parameter beta is set for each semantic mode during training to perform weighted normalization on counting, and a candidate password dictionary is output according to a probability priority queue during generation. The method is suitable for efficient password guessing in offline password auditing, password strategy evaluation and penetration testing.

The invention discloses a manufacturing equipment intelligent operation monitoring method based on industrial vision, and the method comprises the following steps: S1, collecting a continuous image sequence, and forming a visual semantic sequence; s2, inputting into an improved VMama model, introducing a running rhythm phase vector, executing phase conditional processing, and generating a running structure description; s3, constructing an operation semantic mode space; s4, generating an online operation structure description and an operation semantic evolution trajectory; s5, calculating a structure offset degree, executing accumulation judgment, and generating an operation abnormity identifier; s6, reversibility judgment is executed, and an operation risk level is generated; and S7, the operation semantic mode space is updated, and self-adaptive updating of the intelligent operation monitoring capability of the manufacturing equipment is completed. According to the invention, the continuous sensing of the operation state of the manufacturing equipment, the advanced recognition of the abnormal trend and the dynamic evaluation of the risk level are realized, and the technical effects of high monitoring precision, strong adaptability and good operation stability are achieved.