122 results about "Semantic integrity" patented technology

The invention discloses a knowledge question-answering method and system based on topic knowledge graph retrieval enhancement, and the method comprises the steps: firstly extracting a local topic represented in a triple form based on an original document through employing a large language model, carrying out the clustering, and generating a global topic triple set representing the global perspective of the whole document; secondly, on the basis of the global topic triple set, topic-guided entity and relation extraction is adopted, and a mixed knowledge graph is constructed; secondly, providing a semantic perception personalized PageRank algorithm, matching query semantics with semantics of edges in the mixed knowledge graph, and dynamically adjusting the weight of score propagation between nodes; and finally, designing a three-level progressive retrieval mechanism, retrieving multi-level information related to user query from the mixed knowledge graph, and inputting the multi-level information into the large language model to generate a final answer. According to the method, the semantic integrity and retrieval precision of the knowledge graph are remarkably improved, and the accuracy, comprehensiveness and enabling performance of generated answers are ensured.

The invention discloses a text abstract generation method and system based on sparse attention acceleration, and the method comprises the steps: reading long text data, carrying out word segmentation and embedded coding processing, extracting a sequence feature vector, and mapping the sequence feature vector into a query matrix Q, a key matrix K and a value matrix V; constructing an abstract generation network, wherein the abstract generation network comprises a sparse attention calculation module, a feedforward calculation module, a prediction head module and a key value cache module; inputting the query matrix Q, the key matrix K and the value matrix V into an abstract generation network, passing through a backbone network formed by stacking a sparse attention calculation module and a feedforward calculation module for multiple times, processing by a prediction header module, and caching a historical decoding state in real time through a key value caching module to obtain an initial text feature vector; and based on the initial text feature vector, executing an autoregressive decoding process through an abstract generation network, and outputting a final abstract result. According to the method, the decoding process can be accelerated, and the semantic integrity and coherence of the generated abstract are ensured.

The invention discloses a large language model multi-source inference knowledge graph question answering method and system, and the method comprises the steps: extracting entities in a question, linking the entities with knowledge graph entities, and determining a subject entity set; searching paths in the knowledge graph by adopting a graph constraint reasoning method, and obtaining candidate answers and reasoning paths; detecting incomplete path coverage or answer dimension deficiency by using a large language model, triggering question decomposition to generate logically complementary sub-questions, and obtaining answers; establishing an inference source coordination layer, a coordination graph constraint inference source, a planning-retrieval inference source and a sub-problem inference source for multi-source evidence; and standardizing the output of each inference source into a structured evidence triple, sorting and selecting the first K evidences according to relevance, and generating a final answer by utilizing a large language model to carry out inductive inference. According to the method, reasoning can be dynamically adjusted, multi-source evidences can be effectively aggregated, logic consistency and semantic integrity are guaranteed, and the integrity and accuracy of knowledge graph multi-hop questions and answers are improved.

The invention provides a sound wave signal self-supervised learning enhanced robust noise reduction processing method and system, and relates to the technical field of signal processing, and the method comprises the steps: carrying out the feature enhancement of an initial time-frequency representation through a dynamic adaptive mask strategy, constructing a self-supervised reconstruction task based on the mask time-frequency representation, separating noise and signal subspaces in a semantic manifold space, and carrying out the self-supervised learning enhanced robust noise reduction. And establishing a probability propagation network in combination with time domain continuity characteristics to model a local dependency relationship, and finally generating a noise reduction weight and realizing semantic fidelity optimization. The method can effectively improve the noise reduction effect and semantic integrity of sound wave signals in a noise complex environment.

The invention discloses an image fusion method and system based on multi-semantic guidance and mixed experts. The problems that in the prior art, robustness is insufficient, and visual fidelity and semantic integrity are difficult to balance are effectively solved. According to the method, infrared and visible light images to be fused and task identifiers are obtained, and firstly, a CLIP network is utilized to extract high-level semantic features as guide vectors; and then inputting the image and the guide vector into a pre-trained hybrid expert (MoE) fusion network. According to the network, intermediate features are extracted through an encoder, a gating network dynamically activates part of expert subnets according to task identifiers and semantic vectors and calculates routing weights, adaptive nonlinear transformation and weighted fusion are carried out on the features, and finally a high-quality fusion image is reconstructed through a decoder. The method can adapt to different task requirements, and the calculation efficiency is remarkably improved while the image fusion quality and the semantic consistency are improved.

The invention belongs to the technical field of intelligent document processing, and particularly relates to a large model information extraction and structure restoration system for a long text document. The system comprises an extraction point defining and text preprocessing module which is used for carrying out deep preprocessing and analysis on an input unstructured document, extracting text and coordinate information and identifying and protecting a special structure; meanwhile, an intelligent dynamic blocking strategy based on semantic boundaries is adopted, a dynamic overlapping mechanism is combined, and an input document is divided into text blocks keeping semantic integrity; the high-concurrency processing and scheduling control module is used for processing the text blocks in batches, calling a large language model to carry out distributed reasoning and outputting a dispersion result; and the extraction reasoning and result fusion module is used for performing semantic deduplication, entity alignment and confidence fusion on the dispersion result returned by the large language model to generate globally consistent structured output. The method has the advantages of being high in precision, high in speed, controllable in cost and high in robustness.

The invention discloses an Internet webpage content feature extraction method based on artificial intelligence, which comprises the following steps: S1, acquiring a webpage HTML source file and a rendering image, and preprocessing the webpage HTML source file and the rendering image; s2, performing semantic classification on DOM nodes, encoding the DOM nodes into three types of identifiers, and constructing a node label sequence; s3, performing time sequence synchronization on the node attribute vector, the node tag sequence and the visual area set, and performing block-level slicing; s4, inputting the block-level slices into a gated recursive attention network, extracting multi-modal joint representation, and executing attention aggregation; s5, performing structure alignment on the feature fusion sequence, calculating a cross-node consistency distance, and screening a target slice set with high structure cohesion; and S6, mapping the target slice set to a content feature space, and generating a content feature tag set. According to the method, the structural accuracy, the semantic integrity and the multi-modal fusion precision of webpage content extraction are improved.

The invention provides a multi-mode voice interaction method and device and intelligent equipment, is suitable for the technical field of intelligent voice interaction, is applied to the intelligent equipment, and comprises the following steps: in response to a detected voice activity, extracting first voice data in the voice activity, and obtaining video data shot synchronously with the first voice data, a plurality of different users are shot in the video data. And screening out a target user sending the first voice data from the video data. And performing semantic integrity analysis on the text content corresponding to the first voice data. And when the semantic integrity analysis result is that the semantics is incomplete, continuously acquiring second voice data of the target user for a plurality of times, and generating a corresponding target statement with complete semantics after the first voice data and the second voice data are combined. Generating reply data according to the target statement, and outputting the reply data through voice. According to the embodiment of the invention, accurate, coherent and real-time voice interaction with the user needing interaction can be realized.

The invention discloses a document processing method and related equipment, and belongs to the technical field of data processing.The method comprises the steps that in response to a document input instruction, a to-be-processed target document is obtained; performing structure analysis on a to-be-processed target document to generate a directory tree of the target document; calling a dynamic recursive slicing algorithm, traversing and analyzing the directory tree of the target document, and generating a plurality of text slices with dynamic lengths; and vectorizing the plurality of text slices with the dynamic lengths, and uploading and storing the vectorized text slices to a database of a retrieval enhancement generation system. By constructing the directory tree and traversing the directory tree by adopting the dynamic recursive slicing algorithm, the document structure can be more accurately understood, the semantic integrity of the document slices is improved, and the retrieval accuracy and the answer generation quality of the retrieval enhancement generation system are further effectively improved.

This application provides a code digest generation method, model, apparatus, medium, and device, relating to the field of artificial intelligence technology. The method includes: determining an application programming interface (API) information graph corresponding to the code to be processed, wherein the node types in the API information graph include: a first type, and at least one of a second type and a third type, wherein the associated information of the first type of nodes is the application programming interface, the associated information of the second type of nodes is variables, and the associated information of the third type of nodes is API description information; encoding the API information graph to obtain API encoding features of the code to be processed; and determining a code digest of the code to be processed based on the API encoding features and target features of the code to be processed, wherein the target features include: source code encoding features corresponding to the code to be processed, or source code encoding features and AST encoding features corresponding to the code to be processed. This application embodiment can improve the semantic integrity of the code digest.

The application discloses a speech translation method and system and a storage medium, relates to the technical field of speech processing, and comprises the following steps: acquiring a real-time audio stream of source audio, and incrementally generating source language text corresponding to the real-time audio stream; detecting the semantic integrity of the generated source language text in real time; in the case where the semantic integrity is greater than or equal to the integrity threshold value corresponding to a minimum semantic unit, taking the generated source language text as a source language text block; determining target translation corresponding to the source language text block, and outputting the target translation. After the source language text is incrementally generated, the translation of the source language text block and the output are driven based on the semantic integrity, online segmentation translation of long sentences is realized, the translation waiting time is shortened, and the target translation and the original sound are synchronized.

The invention relates to the technical field of semantic retrieval, and discloses an automatic vectorization and semantic retrieval method for business data, which comprises the following steps of: data change capture: sensing newly adding, modifying or deleting operation of data in a business system, extracting business records and binding record identifiers; multi-dimensional quality evaluation: carrying out five-dimensional quality evaluation of integrity, consistency, timeliness, information entropy and security on the extracted service data through an MDQA algorithm, and only allowing the quality standard data to enter a subsequent process; according to the method, through set multi-dimensional quality evaluation intelligent semantic segmentation and incremental vector updating, the retrieval quality is remarkably improved, 22% of low-quality data is filtered through the MDQA algorithm, the retrieval precision is improved by 19.9%, and the user satisfaction degree is improved by 28.1%; the SBAS algorithm ensures segmentation semantic integrity, the recall rate is improved by 25.8%, and retrieval deviation caused by context loss is avoided.

The invention discloses a PDF self-adaptive partitioning method and device, equipment, a medium and a product, and relates to the technical field of data processing, and the method comprises the steps: carrying out the structure reconstruction of an analyzed original PDF document flow, and obtaining a reconstructed PDF document flow; identifying special semantic units, titles and guiding keywords of the reconstructed PDF document flow, integrally replacing the identified special semantic units, judging the continuity of the identified titles, determining segmentation priorities of the titles according to judgment results, and performing segmentation on the titles according to the segmentation priorities. Carrying out semantic binding on the guiding keyword meeting the requirement and the subsequent text to obtain a to-be-segmented PDF document stream; according to the method, the to-be-segmented PDF document stream is subjected to self-adaptive segmentation according to the segmentation priority by adopting the preset multi-level separators, the preliminary segmentation result is obtained, and compared with fixed-length segmentation, the coherence and semantic integrity of the document after PDF segmentation can be guaranteed.

The invention relates to the technical field of computers, and discloses a table metadata processing method and system, electronic equipment and a storage medium, and the method comprises the following steps: carrying out regional semantic segmentation through weighted fusion based on frame geometric features, character coordinates and language features of a table; constructing a structural feature vector based on the row number, the column number and the row list header cell style of the table, performing geometric semantic integrity reconstruction, executing element standardization and semantic extraction, and performing geometric semantic integrity reconstruction on the basis of a region coding library, a unit conversion chain and a knowledge element rule library by utilizing normalized index names, time and region information. And calculating a regional code, a standard measurement value, a knowledge element ID and a retrieval field, matching the normalized index with an existing index system, and outputting a structured record. According to the method, the requirements of high accuracy and traceable structuralization during data processing in a big data environment can be met.

Embodiments of the present application provide a data storage system and method, the data storage system comprises a data input end, a master node, at least two slave nodes, a first database and a second database, the master node and the at least two slave nodes are communicatively connected, and each of the slave nodes is communicatively connected, the master node is used for reading initial data from the data input end and sending the initial data to the at least two slave nodes, the at least two slave nodes are used for calculating the initial data according to a preset data processing flow to obtain device data and metadata, the master node is further used for receiving the metadata and storing the metadata in a mode into a HIVE table of the first database, and the at least two slave nodes are further used for storing the device data into the second database, the metadata calculated is stored in a mode, the problems of low data storage efficiency and high cost caused by using a JSON format for storage are avoided, and the semantic integrity of the initial data is ensured, and the readability of the data is not affected.

This invention discloses a speech interruption decision-making method and system based on multi-granularity semantic integrity prediction, belonging to the field of speech interaction technology. The method includes: extracting multi-granularity semantic features and combining them into a current multi-granularity semantic feature vector; calculating the offset of the current multi-granularity semantic feature vector with a standard semantic pause model to obtain a semantic offset feature vector; performing similarity matching and weighted voting between the semantic offset feature vector and a historical interruption decision case library to obtain the current predicted interruption confidence; and generating an interruption command when the current predicted interruption confidence exceeds a dynamic threshold. This invention solves the problems of existing technologies where speech interruption decisions rely on simple energy thresholds or semantic integrity probabilities, lack utilization of user expression habits and contextual experience, and have low decision accuracy by introducing a standard semantic pause model and a historical case analogy reasoning mechanism, thus achieving more intelligent and accurate speech interruption judgment.

The invention provides a subtitle intelligent batch translation method based on semantic perception. The subtitle intelligent batch translation method comprises the steps that S1, a semantic boundary recognition algorithm with a multi-dimensional batch rule is constructed according to semantic perception logic; s2, identifying a segmentation boundary of the caption text by using a semantic boundary identification algorithm; s3, setting constraint conditions according to the content difficulty, the minimum threshold value, the maximum threshold value and the context semantic integrity of the segmentation boundary of the current batch; dynamically adjusting the segmentation boundary according to a constraint condition; s4, constructing a hierarchical Prompt which comprises a role constraint layer, a term constraint layer, a high-value context layer and a task layer; s5, guiding the LLM model to complete translation of the current batch according to the hierarchical Prompt, screening out high-value information in the current batch, and reserving the high-value information in the next batch; and S6, repeatedly executing the steps S2-S5 until translation of all batches is completed. According to the method, the effect of greatly improving the quality and efficiency of subtitle translation is achieved by accurately sensing the semantic boundary, dynamically and reasonably batching and optimizing translation output layer by layer.

The invention provides a dialogue processing method and device and computer equipment, and the method comprises the steps: determining first information based on current input information; cutting the first information by using a cutting model to obtain second information, the second information representing key information of the first information, and the second information serving as input of a preset dialogue model to generate reply information corresponding to the current input information; wherein the cutting model is obtained by performing reinforcement learning training by using a preset multi-dimensional reward function, and the multi-dimensional reward function is used for updating parameters of the to-be-trained cutting model at least once. In this way, the first information is clipped through the clipping model trained by using a multi-dimensional reward function and a reinforcement learning feedback mechanism, so that the output second information keeps key information, and meanwhile, the semantic integrity and the length controllability are considered.

The invention discloses a semantic repair retrieval enhanced question and answer method, device and equipment. And determining each candidate content combination based on each related content segment and a hierarchical logical relationship among the content segments in a pre-stored reference document, performing semantic repair on each related content segment based on the hierarchical logical relationship of the reference document, and ensuring that the content segments subjected to semantic repair have logical relevance. And semantic fracture or understanding deviation caused by fragmented information is avoided. The preset restoration condition comprises that the total length of the total content segments in the candidate content combination is smaller than a preset length threshold value, and the proportion of each related content segment in the candidate content combination in the total content segments reaches a preset proportion threshold value, and the content segments after semantic restoration can be screened through the preset restoration condition. The waste of computing resources caused by overlong full content fragments is avoided, and excessive interference of redundant information is avoided, so that the semantic integrity and the retrieval precision are balanced.

The application discloses a kind of intelligent structured processing method and system of rules and regulations document, belong to artificial intelligence and natural language processing technical field.Along with the problem that the semantic of long regulation document in prior art is split due to the input length limit of large language model, output is unreliable, the application fuses the paragraph type by small model and semantic feature recognition, corrects label in combination with business rules, and aggregates semantic complete text fragment according to regulation level logic;The segment is constructed as guide prompt input large language model to generate structured result along with type label, and JSON format and field integrity check are set, and when check fails, it is returned to small model result.The method guarantees the integrity of clause-level semantics, improves the accuracy of key field extraction, and ensures system robustness.It is suitable for enterprise compliance management, regulation knowledge base construction and other scenarios, to realize the reliable conversion of unstructured regulation document to unified structured data.

The application discloses a kind of fusion knowledge enhancement and the method and system of original meaning prediction of confrontation training, it is related to natural language processing technical field, obtain target word and its corresponding original dictionary explanation;Original dictionary explanation is carried out semantic integrity evaluation using semantic perception selective enhancement intelligent agent;Target word, original dictionary explanation and supplementary context are structured as structured input sequence;Structured input sequence is sent into original meaning encoder, semantic features are extracted, and original meaning prediction vector is obtained;Using the method based on fast gradient, dynamic confrontation disturbance is applied to word embedding layer during model training process, clean sample loss and confrontation sample loss are optimized jointly to update model parameters;The original meaning label set corresponding to target word is output.The application introduces two big mechanisms of selective knowledge enhancement and confrontation training, and systematically solves the problem that decision boundary is weak due to static dictionary information bottleneck and original meaning long tail distribution in traditional original meaning prediction.

The application provides a subtitle intelligent batch translation method based on semantic perception, comprising the following steps: S1: constructing a semantic boundary recognition algorithm with multi-dimensional batch rules according to semantic perception logic; S2: identifying the segmentation boundary of the subtitle text by using the semantic boundary recognition algorithm; S3: setting constraint conditions according to the content difficulty of the current batch, the minimum threshold, the maximum threshold and the context semantic integrity of the segmentation boundary; dynamically adjusting the segmentation boundary according to the constraint conditions; S4: constructing a hierarchical Prompt, including a role constraint layer, a term constraint layer, a high-value context layer and a task layer; S5: guiding the LLM model to complete the translation of the current batch according to the hierarchical Prompt, and screening out the high-value information in the current batch to the next batch; S6: repeating steps S2-S5 until all batches are completed. The application can accurately perceive the semantic boundary, dynamically and reasonably divide the batches, and optimize the translation output layer by layer, so as to greatly improve the quality and efficiency of the subtitle translation.

The application belongs to the technical field of artificial intelligence video analysis, and relates to a video intelligent description method and system based on size model collaborative fusion and a readable storage medium, which comprises the following steps: extracting key frames from an input video; inputting visual features of the video into a lightweight pre-training model to obtain global feature encoding of the video and a coarse-grained summary of the video; inputting the key frames into a large model to generate fine-grained behavior and interaction detail encoding of the video; setting a prompt word for size model collaborative fusion; inputting the prompt word for size model collaborative fusion, the global feature encoding and the fine-grained behavior and interaction detail encoding into the large model to generate video natural language description with macroscopic integrity and microscopic accuracy. The application has high precision and good integrity in generating multi-granularity semantic description content, and ensures the overall summary of the video content; effectively solves the trade-off contradiction between generalization and detail fidelity of a single model, and significantly improves semantic integrity and scene adaptability.

This invention discloses a text summarization method and system based on sparse attention acceleration, comprising: reading long text data, performing word segmentation and embedding encoding processing, extracting sequence feature vectors, and mapping them to a query matrix Q, a key matrix K, and a value matrix V; constructing a summarization generation network, including a sparse attention calculation module, a feedforward calculation module, a prediction head module, and a key-value caching module; inputting the query matrix Q, key matrix K, and value matrix V into the summarization generation network, first passing through a backbone network composed of multiple stacks of the sparse attention calculation module and the feedforward calculation module, and then being processed by the prediction head module, while the key-value caching module caches historical decoding states in real time to obtain an initial text feature vector; based on the initial text feature vector, performing an autoregressive decoding process through the summarization generation network to output the final summary result. This invention can accelerate the decoding process and ensure the semantic integrity and coherence of the generated summary.

This invention provides an automatic event handling method, apparatus, and system based on semantic analysis, relating to the field of semantic analysis technology. The method includes: acquiring reported information; determining semantic integrity features, event urgency features, and reporting time features based on the reported information; determining recall priority parameters based on the semantic integrity features and event urgency features; determining a correction strategy based on the relationship between the semantic integrity features and a integrity threshold; determining coarse-ranking priority parameters based on the correction strategy and the recall priority parameters; determining target priority parameters based on the coarse-ranking priority parameters and the reporting time features; and determining an event handling sequence based on the target priority parameters. This method introduces a comprehensive judgment of the semantic integrity and accuracy of the reported information, overcoming the shortcomings of related technologies that rely solely on urgency assessment, thereby improving the accuracy and comprehensiveness of information processing.

The application discloses a retrieval enhancement generation method and related products, which comprises the following steps: obtaining a query vector; matching the query vector with a multi-dimensional feature graph to determine a target node; the multi-dimensional feature graph comprises multiple nodes and multiple edges; the node represents a node vector; the edge represents a relationship vector; there is an index relationship between the node vector and the relationship vector; the multiple nodes comprise the target node; performing feature extraction on the query vector to obtain a relationship feature corresponding to the query vector; performing dynamic graph walking in the multi-dimensional feature graph based on the target node and the relationship feature to obtain one or more effective paths; and obtaining a final answer based on the one or more effective paths. The application effectively avoids the "illusion" risk of a large model, ensures that the output content is compliant, reliable and auditable, improves the accuracy of the retrieval result in retrieval enhancement generation, and effectively breaks through the inherent limitations in semantic integrity, multi-hop reasoning capability and relationship intention understanding.

This invention discloses a backdoor attack method, system, and medium based on multi-object semantic coexistence in text-to-image diffusion models. This method utilizes the perspective of multi-object semantic coexistence in text-to-image diffusion models to develop MOBA backdoor attack schemes. First, by constructing a trigger alignment dataset and optimizing backdoor implantation and semantic preservation in parallel, the attack success rate and visual concealment are effectively improved, reducing the impact of semantic corruption on attack effectiveness. Second, during model training, an attention-based decoupling backdoor enhancement mechanism is adopted. By decoupling the attention regions of different objects, the semantic integrity of the input prompt is maintained while the backdoor is activated and the attention distribution is reasonably adjusted to reduce the impact of generation bias on attack concealment. The method of this invention ensures both visual concealment and model performance under benign input while achieving efficient backdoor attacks.

The invention discloses a dynamic switching type intelligent question answering method and system for agricultural product quality control. The method comprises the following steps: firstly, receiving an initial query of a user, and entering an inquiry mode; in the inquiry mode, the system analyzes the semantic integrity of the inquiry, and when the information is incomplete, the user is actively guided to complement key information through multiple rounds of dialogues. And when the information meets a preset integrity condition, the system is dynamically switched to an answer mode. And in an answer mode, the system constructs refined query, performs conversation enhanced retrieval from the domain knowledge base in combination with a complete conversation history, and finally generates a professional solution containing a reasoning process and source reference by a large language model subjected to reasoning fine tuning. The problem that an existing question answering system is difficult to process fuzzy and complex questions in the agricultural field is solved, combination of guiding type intention obtaining and knowledge-driven reasoning is achieved through dynamic switching between an inquiry mode and an answer mode, and the accuracy, specialty and credibility of question answering are improved.

The invention provides a real-time AI intelligent customer service interaction method and equipment with interrupted input and re-splicing. The method comprises the following steps: monitoring and recording a final text fragment output by an automatic voice recognition module and the arrival time of the final text fragment; calculating the time difference of arrival between the current fragment and the previous fragment, and discarding the current fragment or executing a merging strategy; according to a preset merging strategy, splicing the previous text segment and the current text segment to generate a merged text, triggering a revocation instruction to revoke the language model request for the previous text segment, and sending the merged text to the LLM to generate a new reply; and monitoring the times of triggering the merging strategy in the throttling time window through a throttling control mechanism, pausing execution of the merging strategy, and replying according to the arrival time sequence of the final text fragments. The time difference of two ASR results can be automatically detected and combined into the same sentence, and the previous round of LLM request is revoked at the same time, so that semantic integrity and timely response are ensured.

The application provides a nested JSON parent field inheritance flattening extraction method based on path mapping. By pre-configuring the mapping relationship of "parent path and target output field", during the traversal of the JSON tree, the parent node field is automatically identified and extracted based on the current traversal path, and it is written into the inheritance context. When the target extraction node is reached, the target node field and the inheritance context are merged according to the preset conflict processing strategy to generate a stable structure of the flattened output record, and finally a standardized result set output is formed. By configuration instead of hard coding, flexible definition and runtime automatic inheritance of parent-child field relationship are realized. Compared with existing solutions, the data semantic integrity and consistency are significantly improved, the development and maintenance cost is greatly reduced, and the method has strong universality and engineering landing value, and can be widely applied to API data access, ETL offline processing, real-time data middle platform, data warehouse construction and the like.