328 results about "Semantic representation" patented technology

The application discloses a government affair text auxiliary writing system and method based on a multi-modal large model, and relates to the technical field of artificial intelligence and informationization; comprising: step 1: converting heterogeneous government affair data into unified semantic representation, realizing cross-modal deep alignment through fine-grained contrast learning; encoding text, image and voice features by using an encoder respectively, and constructing a training set containing 100,000 pairs of government affair image-text samples; weighting and fusing the text, image and voice features, dynamically adjusting the weight through an attention mechanism, training a large model, step 2: performing lightweight deployment of the large model, and step 3: using the large model to understand and identify user intention, and outputting government affair text.

This application discloses a method and system for multilingual enhancement based on a large language model using model ensemble. The system includes: a pre-trained multilingual translation model, a semantic representation mapping module, and a large language model. The multilingual translation model is used for multilingual semantic modeling and language generation, including a multilingual encoder module and a multilingual decoder module. The semantic representation mapping module is used to transform the latent space representations of different models into an interactive unified semantic space based on a cross-model representation mapping mechanism. The output of the multilingual encoder is mapped to the unified semantic representation space of the large language model, and the mapped semantics are input into the large language model to perform language-independent instruction understanding. The intermediate semantic representation output by the large language model is mapped and transformed to a cross-attention representation space, generating the final output text under the target language distribution. The system of this application outperforms existing technologies in terms of efficiency, stability, and generation quality in multilingual capability extension.

The embodiment of the application discloses an automatic online quality inspection method and device based on AI image recognition, a medium and a product; the method comprises the following steps: in the automatic script execution process, the application interface screenshot is collected at the preset node, and the context metadata corresponding to the screenshot is obtained; the screenshot and the context metadata are input into a multimodal model to obtain a structured semantic representation for the screenshot, wherein the structured semantic representation at least includes type information, text content, spatial position information and hierarchical relationship of interface elements; based on the structured semantic representation, the interface elements are uniformly semantically modeled to form a standardized representation; a rule engine is called to perform rule matching calculation on the standardized representation to obtain a matching result, a deviation value and a confidence; according to the matching result, the deviation value and the confidence, an inspection judgment result is generated, and structured evidence information containing a rule hit path is output.

This specification provides a method and apparatus for generating user-described text based on a text generation network. The method includes: inputting various features of a target user into a first encoder; obtaining initial user feature vectors corresponding to each feature through the first encoder; encoding these features based on a self-attention mechanism to obtain an encoded state vector; inputting the encoded state vector into a retrieval model; retrieving K sentences from a human-made knowledge base through the retrieval model; determining the character encoding vectors corresponding to each character contained in the K sentences; determining attention coefficients based on the decoder's output feedback vector and the character encoding vectors; and weighting and summing the character encoding vectors based on the attention coefficients to obtain a semantic representation vector; inputting the encoded state vector and the semantic representation vector into a decoder; and generating user-described text for the target user through the decoder, with the decoder's hidden state serving as the output feedback vector. This method can improve the quality of the obtained text.

The application discloses a COBOL-to-Java program conversion method and system based on a syntax tree, belongs to the technical field of computer program language conversion, and comprises the following steps: obtaining a COBOL source program file to be converted and configuration rules; performing lexical analysis and syntax analysis to generate an abstract syntax tree; converting COBOL syntax elements into corresponding Java semantic representations according to pre-defined semantic mapping rules; establishing an accurate access chain of a main program and COPYBOOK internal variables through variable path tracking; detecting and solving naming conflicts of COBOL different-level same-name variables based on a scope tree, and renaming variables by adding a parent class name prefix; and generating the converted Java semantic representations into a Java source code file. The application realizes high-fidelity and high-automation COBOL-to-Java conversion, significantly reduces the migration cost of a legacy system, and generates code with strong readability and excellent performance.

This disclosure proposes a dialogue processing method, a training method for a question rewriting model, and a device thereof. The dialogue processing method includes: when information is missing in the user question, inputting the user question and historical dialogue content into the encoding layer of the question rewriting model to obtain the semantic representation vectors of each word in the historical dialogue content and the semantic representation vectors of each word in the user question, as well as the keyword scores of each word in the historical dialogue content; then, the decoding layer accurately selects keywords semantically related to the user question from the historical dialogue content based on the semantic representation vectors of each word in the user question, the semantic representation vectors of each word in the historical dialogue content, and the keyword scores, and explicitly generates a fully rewritten user question based on these keywords. This accurately supplements the missing information in the user question, achieving completeness and readability of the user question content and improving the user experience.

The work ticket intelligent checking method based on multi-modal large model knowledge graph and large model reasoning belongs to the field of power system application technology, first extracts text feature vectors and image feature vectors through an OCR module and an image encoder, forms a comprehensive semantic representation after multi-modal fusion. Extract entities and relationships from power safety regulations and historical operation data to generate a power safety knowledge graph. Check the format and integrity of the basic fields on the ticket, use LLM to combine context and graph nodes for semantic reasoning to generate semantic judgment results. Fuse three layers of results to calculate a comprehensive checking score and output an intelligent determination result. The present application introduces multi-modal input analysis, knowledge graph construction and large model semantic reasoning technology to construct a structured, interpretable and iteratively optimized intelligent work ticket checking system that realizes the full-process intelligentization from ticket image recognition to semantic understanding and from explicit rule checking to implicit logical reasoning.

This invention discloses a lightweight method for detecting foreign objects (FOOs) in power transmission lines, relating to the fields of power line inspection and computer vision. A YOLOv10-based FEO detection model is constructed, comprising a feature extraction part, a feature interaction part, and a detection head part. The C2f module in the feature extraction and feature interaction parts is replaced with an expert dynamic extraction module, and the detection module in the detection head part is replaced with a parameter-sharing detection module. The trained FEO detection model is obtained by collecting and evaluating a calibrated FEO dataset of power transmission lines. FEO keyframes are input into the trained FEO detection model to obtain detection results. The expert dynamic extraction module acquires feature maps from multiple receptive fields, achieving rich representation in FEO detection environments with varying scales. The parameter-sharing detection module, based on this, reduces redundant computation and enhances the semantic representation of small targets through parameter sharing and grouping normalization, effectively improving the efficiency of FEO detection in power transmission lines.

This application relates to the fields of autonomous driving scene representation and artificial intelligence technology, and particularly to a method for causal saliency mining and anti-interference semantic representation in autonomous driving scenarios. The method includes the following steps: encoding multimodal scene elements separately and concatenating the encoding results to obtain a unified scene representation; inputting the unified scene representation into a contrastive representation module, and performing global importance ranking based on the internal weights of a self-attention mechanism; classifying scene elements into different categories and filtering the categories to construct category-aware automated positive and negative samples; performing context alignment on the category-aware automated positive and negative samples based on a recoding mechanism; and constructing a contrastive loss function to enable the model to distinguish key causal information from interference noise in the feature space. This application, through an attention-guided automated sampling mechanism and recoding strategy, filters out key features with causal relationships from high-dimensional heterogeneous driving scenarios and suppresses spurious correlation noise in the environment.

This application discloses a multimodal temporal causal intelligent decision-making method, apparatus, device, and medium for complex scenarios, relating to the field of artificial intelligence technology. The method includes: constructing an integrated intelligent decision-making framework from multimodal data; achieving shared semantic representation of multi-source data through unified encoding; obtaining reliable fusion causal strength through dual-channel causal discovery; constructing and dynamically updating a causal knowledge graph based on this strength; and generating compliant decisions by combining constraint projection and counterfactual verification. The beneficial effects are: realizing dynamic causal knowledge updating and generating constraint-compliant and reliable decisions based on multimodal data; ensuring that decisions strictly comply with physical and business rules; and improving the reliability, robustness, and real-time performance of decisions in complex scenarios.

The application belongs to the field of natural language processing, and particularly relates to a method and system for paraphrase recognition based on semantic primitive knowledge and abstract semantic representation, which comprises the following steps: performing word segmentation on a sentence, and performing word-level vector representation and semantic primitive knowledge representation; performing mean value processing on the semantic primitive knowledge representation result, and extracting interactive attention feature information of the mean value processing result by using global semantic information to obtain global semantic primitive representation; performing abstract semantic analysis on a to-be-recognized paraphrase sentence from a sentence structure to obtain a single-root directed acyclic graph, and performing global semantic primitive representation and word-level vector representation; extracting global and local feature information in the order of the directed acyclic graph, and performing distance feature measurement on the information; inputting the distance feature measurement result into a neural network to obtain a recognition result; the application introduces external semantic primitive knowledge to perform semantic representation, the accuracy of the semantic primitive knowledge representation is assisted by global semantic information, and the abstract semantics of a Chinese paraphrase sentence is analyzed to obtain semantic relations, so that the accuracy of paraphrase recognition is improved.

The application discloses a patent theme graph generation method and system based on deep semantic hierarchical clustering, comprising: obtaining a patent literature set PT to be generated into a patent theme graph; using a patent deep semantic representation model to perform semantic coding on each patent literature in the patent literature set PT, and obtaining a semantic representation vector matrix V={v1,...,v N}; inputting the semantic representation vector matrix V into a hierarchical clustering algorithm to obtain a hierarchical clustering tree structure corresponding to the patent literature set PT; generating a corresponding theme description for each non-leaf node on the hierarchical clustering tree structure; and combining the generated theme description and the hierarchical clustering tree structure into a patent theme graph with an upper and lower hierarchical structure. The patent theme graph generation method and system can help users quickly mine upper and lower hierarchical relationships of patent literatures at a theme level, meet the analysis needs of users for large-scale patent information, and improve the efficiency of patent analysis of users.

The application provides a large language model training system. The large language model training system comprises a knowledge management module, a training data processing module, a first stage training module, and a second stage training module. The knowledge management module is configured to obtain an audit rule and generate a rule training set based on the audit rule. The training data processing module is configured to obtain a case data set and construct a case training set based on the case data set. The first stage training module is configured to train an initial large language model based on the rule training set to generate a rule large language model. The second stage training module is configured to train the rule large language model based on the case training set to obtain a target large language model. The technical problem that a large language model is difficult to form stable and structured rule semantic representation in parameters in the prior art, and the recognition ability of the trained large language model for implicit and evolving content is insufficient can be solved.

The application discloses a sequence recommendation method based on semantic discretization and multi-interest modeling, and belongs to the field of natural language processing. The method encodes item metadata text into multiple discrete semantic tokens, extracts multiple interest query vectors from user historical comments, and calculates a recommendation score at the semantic level by matching the interest query and the semantic token of the candidate item. Meanwhile, the user historical interaction sequence is used to model collaborative preference, so as to fuse the semantic score and the collaborative score to generate a final recommendation result. Through user feedback aggregation and a multi-query mechanism, the modeling ability of user multi-interest preference is improved. Through fine-grained matching of the user interest query and the discrete semantic representation of the candidate item, the semantic interaction ability between the user and the candidate item is improved.

This invention discloses a method and system for on-site monitoring and early warning based on retrieval enhancement. The method involves acquiring multimodal data from on-site inspections, transforming this data into several independent semantic segments based on its acquisition attributes, preprocessing these segments to obtain on-site state representation vectors, filtering and identifying valid document fields in a document library, performing natural language vector transformation on the valid document fields to obtain knowledge segment vectors, constructing a trusted knowledge segment vector library based on the document library's updates, retrieving matching knowledge segment vectors from the trusted knowledge segment vector library based on the on-site state representation vectors, and performing early warning operations for on-site anomalies based on the semantic relationships between the matching knowledge segment vectors. By forming semantic representation vectors from on-site inspection multimodal data and knowledge segment vectors from a benchmark document library, the method achieves matching retrieval between vectors, resulting in standardized early warning operations.

The application discloses an intelligent learning question and answer analysis method and application system based on government service, relates to the technical field of artificial intelligence and government service, and comprises the following steps: performing semantic analysis on question and answer data submitted by a user, extracting intention features and entity features to form semantic representation, identifying mutual exclusion relations of cross-department rules in a government rule library, and generating a conflict set. On this basis, the time limit for validity and the time limit for completing the matter of the user are combined to calculate the time limit priority, the conflict resolution path is prioritized, and a multi-path decision tree is constructed. Meanwhile, high-frequency resolution paths are learned from historical question and answer data and a knowledge base is established, so that experience reuse is realized. When a new question and answer request is received, the resolution path in the knowledge base is matched preferentially, and when the matching fails, the optimal reply is output through the decision tree, so that the intelligent level of government question and answer and the cross-department business processing efficiency are improved.

The application discloses a kind of commodity representation generation model training method and the method and device of associated commodity identification, it is related to computer technology field.The specific embodiment of the method includes calling commodity representation generation model, to calculate the residual vector of sample commodity based on the semantic information of commodity in sample commodity set, and further determine the semantic representation of sample commodity.By calling commodity representation generation model, to generate the co-occurrence representation of sample commodity based on corresponding user behavior information, finally according to the semantic representation and co-occurrence representation obtained training commodity representation generation model.This embodiment can be combined with the semantic information of commodity to represent the information of commodity, and accurately feature representation is carried out to commodity in combination with the co-occurrence information of commodity, to generate accurate commodity representation data, provide data basis for subsequent identification operation.

The application discloses an audio encoding method, an audio decoding method and an audio encoding and decoding system. The method comprises the following steps: obtaining audio to be transmitted and a transmission bit rate; extracting a semantic feature sequence of the audio, and performing vector quantization on the semantic feature sequence based on a semantic codebook to obtain a discrete semantic token sequence encoded based on a first bit rate; extracting a first acoustic feature of the audio, determining a second bit rate which satisfies a transmission bit rate constraint together with the first bit rate, performing downsampling processing on the first acoustic feature based on the second bit rate to obtain a second acoustic feature, and performing vector quantization on the second acoustic feature based on an acoustic codebook to obtain a discrete acoustic token sequence encoded based on the second bit rate; and transmitting the discrete semantic token sequence and the discrete acoustic token sequence to an audio decoding device to reconstruct the audio. The application solves the technical problem that a traditional audio encoding and decoding scheme is difficult to simultaneously consider strong semantic representation and high-fidelity reconstruction in an ultralow bit rate scene.

The application provides a data center computing power time sequence prediction method and device, and relates to the technical field of resource scheduling. The method comprises the following steps: based on the semantic similarity between a target variable sequence of a target data center at a current time and each historical power sequence in a vector database, retrieving semantic representations of multiple prediction power sequences; performing time sequence dependence and variable association modeling on the target variable sequence and a covariant sequence to extract time sequence dynamic feature representations; and performing weighted fusion of the semantic representations of the multiple prediction power sequences and the time sequence dynamic feature representations through dynamic gating fusion to obtain a computing power prediction value of the target data center at a future specified step. The data center computing power time sequence prediction method and device provided by the application can guarantee high precision while significantly improving the prediction robustness and generalization ability of the model in the face of unknown modes and scene mutations through deep fusion of semantic-level historical experience retrieval and deep time sequence context modeling.

The application discloses a model training method and device and electronic equipment. The method comprises the following steps: acquiring space availability data, embodied visual question answering data and robot demonstration data; constructing a multi-modal training data set based on the space availability data, the embodied visual question answering data and the robot demonstration data; pre-training an initial visual language model using the multi-modal training data set to obtain a target visual language model; adding a motion head to the target visual language model to obtain an initial visual language motion model; screening a robot motion trajectory data set from the robot demonstration data; pre-training the initial visual language motion model using the motion trajectory data set to obtain a target visual language motion model; and the target visual language model in the target visual language motion model is used to obtain a fused semantic representation according to input data, and the motion head in the target visual language motion model is used to output a motion control parameter of a robot according to the fused semantic representation.

This invention relates to the field of land consolidation carbon effect assessment technology, and discloses a system and method for quantitative assessment of carbon effects based on land consolidation type. The system includes a regional semantic layer module, a consolidation project semantic layer module, and a parameter semantic layer module. It also includes a parameter semantic processing module for constructing a structured semantic representation of carbon factor parameters. These carbon factor parameters are associated with an uncertainty descriptor ontology, which encapsulates metadata information characterizing the statistical uncertainty and applicability constraints of the carbon factor parameters. This system and method for quantitative assessment of carbon effects based on land consolidation type establishes a semantic mapping relationship between the regional semantic layer, the consolidation project semantic layer, and the parameter semantic layer through a cross-layer semantic association module. This accurately determines the applicability of parameters, avoids assessment bias caused by parameter misuse, and effectively solves the problems of low coverage and poor regional adaptability of localized parameters.

This invention discloses a container shipping route analysis method based on artificial intelligence analysis, belonging to the field of shipping intelligence technology. The method includes: collecting route operation data during the operation and management of a target container shipping route, performing time alignment and normalization processing to generate an actual route event sequence; embedding route governance rules into corresponding route segments based on the actual route event sequence to construct a route governance semantic representation; outputting the arrival time series of each port of call and the reliability evolution characteristics as the route progresses through an artificial intelligence analysis process based on the route governance semantic representation; and generating a counterfactual route trajectory sequence using the arrival time series and reliability evolution characteristics. This invention reconstructs the theoretical evolution process of a shipping route under given operational constraints by constructing a counterfactual route trajectory with route governance semantics, providing a predictable and alignable expressive basis for the operational state of the route in both time and space dimensions.

This application discloses a method and system for network security situation awareness of a private energy network integrating the Internet of Things (IoT) and artificial intelligence. The method includes: extracting frequency fluctuation feature sequences and communication interaction data based on the energy network's operating frequency signal and the network communication data stream; constructing a unified frequency-time reference axis for the entire network, mapping the communication interaction data to the frequency-time reference axis, and generating a frequency-anchored data sequence; performing semantic parsing on the energy network protocol messages in the communication interaction data, constructing a semantic representation of control commands and their expected physical effects model; performing semantic-temporal consistency verification to obtain semantic-physical coupling residuals; calculating the temporal reliability score of the communication data and identifying abnormal control behaviors; inputting the temporal reliability score and abnormal control behaviors into a preset situation assessment model, and outputting the network security situation value of the private energy network. This application improves the accuracy of network security situation detection.

This invention discloses a multi-scale, multi-semantic collaborative target detection method, system, medium, and product, belonging to the field of target detection. The method comprises: acquiring image data containing the target to be detected; performing multiple iterative feature processing on the object to be processed based on a preset detection model to obtain the final features; in each iteration, extracting initial features from the current object to be processed, grouping them into four sub-features along the channel dimension, and performing texture and edge enhancement, structural prior extraction, local context modeling, and global semantic representation generation through the four branches to obtain the features of each branch; fusing the features of each branch with the initial features to update the object to be processed; the object to be processed in the first iteration is the image data; after a preset number of iterations, performing target detection based on the final features and outputting the results. By implementing this invention, the problem of insufficient perception accuracy and robustness of existing target detection methods for multi-scale and multi-type targets in complex scenes can be solved.

This application provides a method for multi-source intelligence semantic analysis and compliance risk early warning, applied to intelligent processing devices. The method includes: transforming multi-source heterogeneous intelligence data into a unified semantic representation vector through a cross-modal semantic mapping model; performing entity linking and relationship reasoning on the unified semantic representation vector based on an enterprise multi-source intelligence knowledge graph library to generate an intelligence semantic analysis graph; matching the intelligence semantic analysis graph with a multi-level compliance rule library through an enterprise compliance rule reasoning engine to mine and assess the severity and associated transmission paths of potential compliance risk points to generate a risk transmission path diagram; and generating tiered compliance risk early warning information based on the risk transmission path diagram. This application improves the effectiveness of multi-source intelligence semantic analysis and the rationality of compliance risk early warning compared to existing solutions, and better meets the actual needs of enterprises for refined management of compliance risks.

This application provides a method and apparatus for generating traditional Chinese medicine (TCM) prescriptions based on multi-task joint optimization. The method includes: constructing a training set based on historical clinical case data, the training set comprising multiple samples and sample weight labels, each sample containing clinical text features, TCM prescription labels, and prescription set labels; constructing an enhanced training set based on the training set; training a multi-task joint optimization prescription generation model based on the enhanced training set; predicting TCM prescription sequences using an autoregressive approach, with the TCM prescription labels as the supervision signal; a set prediction branch used to predict the presence or absence of each TCM herb in the global TCM dictionary based on the semantic representation of clinical text features, with the prescription set labels as the supervision signal, and introducing sample weight labels into its loss function to enhance attention to low-frequency TCM herbs; training the model by weighted fusion of the loss functions of the above two branches; and generating TCM prescriptions using the trained model.

This invention relates to the field of smart city technology, and in particular to a method and system for accurately pushing dynamic information from smart city bus stop signs. The method involves constructing a query semantic representation structure by collecting query requests and scene state information, obtaining a joint understanding representation through multi-level semantic association modeling with scene state information, performing scene-constrained multi-hop reasoning in a public transport knowledge association network, filtering and generating candidate answer sequences, calculating scene suitability scores, selecting the optimal answer to generate push content, and collecting interactive feedback to adaptively adjust feature fusion weights to form a closed-loop optimization. This achieves accurate dynamic push of public transport information and improves user experience.