64 results about "Generative semantics" patented technology

The invention provides a natural language text data intelligent classification method and system based on deep learning, and relates to the technical field of natural language processing, and the method comprises the steps: 1, employing a context awareness mechanism to analyze the real semantics of a target vocabulary according to an antagonistic variant existing in a text, and obtaining a target vocabulary; in combination with a word meaning library and a pre-training process of a dynamic learning rate adjustment strategy, generating a candidate replacement vocabulary set with consistent semantics; and step 2, based on the candidate replacement vocabulary set, performing multi-dimensional semantic similarity calculation and emotional tendency discrimination, determining applicable vocabularies conforming to an original culture background through a context adaptation strategy, and generating a standardized text sequence. According to the method, through multi-dimensional semantic analysis, cultural context fusion, cross-granularity feature construction and dynamic parameter correction, the accuracy and adaptability of natural language text classification are realized.

The invention discloses a knowledge base technical method and system based on RAG retrieval enhancement, and relates to the technical field of knowledge bases. The method comprises the steps that a source document is analyzed into structured text fragments, and semantic vectors are generated; splitting the text segments into minimum knowledge units, extracting concepts and behavior trigger words to construct a concept association graph, and storing session abstracts in a fixed-length annular structure; the method comprises the following steps: receiving user query, generating a query vector, retrieving a most relevant text fragment, constructing and de-duplicating candidate knowledge units by combining hierarchical diffusion of a concept association graph and an annular abstract matching result, deeply splicing original text paragraphs according to a graph path, generating a dynamic prompt box, calling a generative model to generate a preliminary answer, and executing verification. And the verification state is marked on the final answer. By constructing a concept association map, the activeness weight and the time sequence fingerprint of a map edge are updated in real time, and accurate capture of deep semantics and logical relationships of query intentions is realized.

The invention discloses an algorithm method for large-model long context reasoning, which relates to the technical field of large language models and comprises the following steps of: dividing an input long text sequence into a plurality of initial text blocks; a semantic abstract is generated based on the initial text blocks, clustering analysis is conducted on the semantic abstract, the initial text blocks with similar semantics are combined into semantic hyperblocks, and a context organization structure with semantic representativeness is formed; the method comprises the following steps: generating key value cache data of each token in a large model preprocessing stage, grouping the key value cache data by taking semantic hyperblocks as logic boundaries, and establishing a mapping table for recording storage positions and states of the semantic hyperblocks; on the basis of semantic hyperblocks, semantic representative vectors of the semantic hyperblocks are combined, correlation scores between query vectors and the semantic hyperblocks are calculated, a block importance prediction model learns on the basis of context dependency features, high-order semantic prior is provided for subsequent attention screening, and key information omission caused by position offset is avoided.

The invention discloses a personalized semantic understanding system for assisting an autonomous mobile platform, and the system comprises a user personalized word meaning library which is used for recording the mapping relation between the user habit expression and the corresponding operation; the fuzzy semantic analysis module is used for identifying fuzzy expression in a natural language instruction of a user and generating a plurality of candidate targets or operation options of semantic generalization; the multi-round context modeling module is used for establishing a semantic context based on session history and performing disambiguation and complementation on current semantics; and the intention generation module is used for converting the processed semantic information into a structured instruction. The personalized semantic understanding system is reasonable in structural design, all the modules cooperate with one another, the effect that 1 + 1 is larger than 2 can be achieved, fuzzy semantic understanding and multi-round context modeling can be conducted, the personalized requirement of a user can be met, semantic understanding precision is high, user adaptability is good, and the personalized semantic understanding system is suitable for man-machine interaction of an autonomous mobile platform.

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for generating an audio signal from input text. In one aspect, a method includes receiving a request to convert an input text to an audio signal, where the input text includes a plurality of lexical meta text inputs; generating, using a first generative neural network, a semantic representation of the lexical text input, the semantic representation comprising semantic terms representing semantic content of the lexical text input, each semantic term selected from a word list of semantic terms; generating an acoustic representation of the semantic representation using a second generative neural network and on at least the semantic representation, the acoustic representation comprising one or more respective acoustic lems representing acoustic properties of the audio signal; and processing the acoustic representation using a decoder neural network to generate an audio signal.

The invention discloses a question and answer interaction method, device and equipment based on artificial intelligence and a medium, and relates to the technical field of artificial intelligence in the professional service fields of finance, insurance, medical treatment, banking and the like, and the method comprises the steps: segmenting a historical dialogue into segments with coherent themes, and constructing the segments into memory units; generating a semantic embedding vector and a time decay weight for each unit to form a memory bank; retrieving related memory units from a memory bank based on user query, comprehensive semantic relevance, theme consistency and time decay weight; the query and related units are combined into a context input generative language model, and a topic consistency constraint is applied during the generation process to generate a topic consistent reply. By optimizing the memory granularity and enhancing the memory representation and timeliness, multi-dimensional accurate retrieval and constraint generation are realized, and the system improves the accuracy, coherence and service consistency of question and answer interaction.

The invention relates to the technical field of intelligent interaction, in particular to an artificial intelligence interaction robot, which comprises a statement recognition module for analyzing a statement structure, separating trunk and affiliated contents, marking a grammar role to generate a semantic structure, a time marking module for recognizing and marking a time sequence, and a path construction module for extracting a historical path for rearrangement. The expression cleaning module identifies and clears redundancy, and the output generation module is connected with the buffer area and the path to generate complete output. According to the method, trunk and affiliated components are identified through multi-round statement structure analysis, accurate separation and reconstruction of semantic elements are realized in combination with word role marking, a time arrangement list is generated according to a time expression sequence, semantic evolution judgment is improved, and a historical path is matched to construct an expression rearrangement structure to enhance semantic connection; and identifying a redundant statement optimization structure and definition, and realizing continuity and logic consistency of interaction contents.

The invention provides an AI-based multi-user interactive language learning system and method, and relates to the technical field of interactive language learning. Real-time structured analysis of a multi-user mixed voice stream is realized through a voiceprint separation technology, a grammar error density spectrum is constructed to quantify user grammar error distribution characteristics, and a culture conflict intensity field is established; cross-cultural dialogue conflict intensity is dynamically calibrated, a semantic transition trajectory is generated, a logic fault evolution law is accurately captured, and a traditional single-dimensional static learning mode is broken through; a language adaptation index is dynamically generated on the basis of a weight prediction model, collaborative modulation of a grammar reconstruction task package, a culture tuning script and a semantic bridging task is achieved, culture conflict mediation deeply conforms to real-time conflict intensity, and dialogue fault features are accurately matched through semantic logic repair; through a three-channel task distribution mechanism of directional insertion, multicast broadcast and random allocation, efficient collaboration of personalized grammar training, immersive culture drill and distributed logic repair is ensured.

The invention provides an MDS dimension reduction-based retrieval enhancement generation semantic matching optimization method and system and a storage medium. The method comprises the following steps of: obtaining a user problem and high-dimensional semantic vectors of a plurality of candidate text segments; constructing a distance matrix based on the two high-dimensional semantic vectors to obtain a high-dimensional semantic similarity; performing dimension reduction on the distance matrix by adopting a multi-dimensional scale analysis (MDS) method, and calculating low-dimensional semantic similarity between the user question and each text fragment based on a distance relationship after dimension reduction; performing weighted fusion on the low-dimensional semantic similarity and the high-dimensional semantic similarity to generate a mixed similarity; and sorting the mixed similarities from high to low, and selecting the text fragments corresponding to the top n of the mixed similarities for generating response output. By reducing the dimension of the high-dimensional retrieval vector to the low-dimensional space, the interference of irrelevant dimensions is reduced, the semantic matching precision and efficiency are improved, and then the quality of the text recalled by the large language model is improved.

The invention discloses a spatial relationship natural language statement conversion method and device taking context into account, and relates to the field of spatial relationship recognition, and the method comprises the following steps: obtaining spatial elements; determining a target spatial relationship description mode matched with the spatial elements, converting the natural language statement into a target normalized structure statement, obtaining context information, and supplementing the context information to the target normalized structure statement; according to the method, the scale context information is queried through the pre-constructed knowledge graph, and the context information is supplemented to the standardized structure statement by referring to the benchmark constraint context information and the direction path constraint context information, so that the spatial relationship description statement with complete semantics is generated, effective analysis of the hidden context of spatial relationship description is realized, and the semantic description efficiency is improved. The problem that current spatial relationship description research lacks mining of spatial relationship context information is solved, and accurate and efficient positioning of geographic targets is facilitated.

The invention discloses a semantic analysis and recognition method based on artificial intelligence, particularly relates to the field of semantic analysis, and is used for solving the problems of frequent semantic offset and inconsistent translation results caused by difficulty in accurate recognition and disambiguation of cultural load words in existing cross-language text processing. The method comprises the following steps: performing text structure decomposition and culture metaphor analysis on a cross-language contrast corpus, identifying culture load words with specific culture meanings, further obtaining semantic vector sets of the culture load words in a source culture context and a target culture context, and calculating a semantic escape distance between the culture load words and the target culture context to generate an escape path; and training a context disambiguation model in combination with the escape path and the context, so that the model can output final semantic probability distribution of the culture load words in the target culture when the to-be-analyzed text is input, and generates a semantic mapping prompt in a cross-language semantic conversion process based on the probability distribution. Therefore, cross-culture accurate semantic analysis and prompt of the culture load words are realized.

The invention relates to a self-adaptive voice semantic communication method based on hierarchy time sequence importance, which belongs to the field of voice analysis, and comprises the following steps: a sending end performs priority division on a discrete feature matrix according to hierarchy and time sequence attributes of voice features, and screens the features according to importance scores in a feature matrix packaging and selecting link; the method comprises the following steps: interacting with wireless communication through a channel adaptive scheduling module, introducing a channel feedback mechanism, and dynamically adjusting transmission power and a resource allocation strategy by sensing current channel state information; and after completing signal demodulation, a receiving end inputs the acquired sparse feature flow into a voice restoration module, and globally reconstructs the received features by using voice priori knowledge of deep pre-training. According to the method, a hierarchical speech feature extraction technology based on a discrete codebook and a generative semantic repair technology are combined, so that the speech word error rate in a severe channel environment is reduced to the maximum extent, and the semantic intelligibility of a receiving end is improved.

The invention relates to a pediatric depression risk level determination method and device based on a big language model, and the method comprises the steps: carrying out the unified coding, noise elimination, sentence segmentation, part-of-speech tagging, voice-to-text conversion and desensitization processing based on original text data, and obtaining a standardized text; inputting the standardized text data into a preset large language model, and generating a semantic vector through word segmentation embedding; based on the semantic vector, utilizing a preset emotion recognition model to extract a depressive symptom entity and feature items thereof; based on the symptom entity and the feature item, mapping according to a preset depression standard dictionary and a synonym expansion word list to obtain a quantifiable structured feature vector; and based on the structured feature vector, obtaining a risk assessment level of the depression of the children. According to the method, the recognition accuracy of the depressive symptoms in the unstructured text is improved, and the time cost of manual interpretation is reduced.

The present invention discloses a method, system, device and storage medium for generating multilingual rhyming lyrics. A speech generation model is used to capture the pronunciation of words, and then rhyming word pairs are generated, greatly improving the rhyming quality of the generated words. At the same time, an autoencoder model with the rhyme as the starting input is adopted during the generation process, which can generate lyrics with more coherent semantics. Moreover, it supports the generation of multilingual lyrics, making the generated rhyming words more diverse.

This invention provides an automated compiler testing method based on specification inference and loop consistency verification to address the problems of traditional compiler testing, which relies on differential testing, requires reference implementations, and cannot effectively test domain-specific compilers. The method first extracts a syntactic skeleton set and a detailed reference set from heterogeneous specification documents; it then uses a large language model to infer type-safe formal grammar rules through multi-agent collaboration; based on the inferred grammar, it generates semantically valid test cases using a seed-preserving abstract syntax tree-level mutation strategy; it executes a compile-decompile-recompile loop on the test cases and detects compiler defects through binary-centralized comparison and noise normalization. This invention achieves systematic automated testing of compilers in the absence of reference implementations and execution oracles, improving compiler reliability and system security.

The invention relates to a data enhancement method based on a two-way translation adversarial network, and belongs to the field of natural language processing. In low-resource language machine translation, the problem that a machine translation result deviates in a specific field due to the lack of large-scale diversified training corpora is solved. In order to relieve the influence caused by deviation in a specific field, the invention provides the method, and the method combines adversarial training and a bidirectional back translation technology, and generates a high-quality adversarial sample and diversified training data by using a pre-trained shielded language model so as to enhance the robustness of the model. Firstly, adversarial training is used for carrying out data enhancement on a data set, then a bidirectional translation method is used for generating texts with similar semantics but different expressions, then a pre-training shielding language model is used for generating text variants with reasonable semantics, finally, bidirectional translation and MLM are fused to generate more reasonable and diversified enhanced data, and the data are used for retraining the model. According to the method, the robustness and the performance of the NMT model are remarkably enhanced on the data level.

The present invention relates to the field of machine learning, and specifically to a method and device for copyright protection of large language models based on fingerprint member probability offset signals. A fingerprint data set is constructed, and the fingerprint data set is symmetrically divided into a training subset and a reference subset; the original model is fine-tuned and trained using the reference subset to obtain a reference model; according to the training subset, a semantically equivalent perturbation sample set is generated; according to the semantically equivalent perturbation sample set, the probability offset indexes of the model to be verified and the reference model are calculated respectively; according to the probability offset indexes of the model to be verified and the reference model respectively, the average signal strength index of the fingerprint data set is determined; according to the average signal strength index, it is judged whether the model to be verified constitutes infringement. The present invention can realize covert, efficient, accurate and highly robust model copyright verification in gray-box or black-box scenarios after the leakage of large language models.

The present application discloses a method and apparatus for training a semantic vector extraction model and representing semantic vectors. The method for training the semantic vector extraction model includes: obtaining a conversation corpus from the same scenario as a downstream task corresponding to the semantic vector as a first pre-training text; performing secondary training on the trained BERT language network based on the first pre-training text to generate a BERT language network that focuses on nouns and verbs in the text; obtaining text classification data and generating a text similarity dataset based on the text classification data; and fine-tuning a multi-task classification network based on the text similarity dataset to generate a semantic vector extraction model.

The invention provides an end-side large language model reasoning method and device, equipment, a medium and a program product. Relates to the technical field of artificial intelligence. The method comprises the following steps: in response to a received to-be-reasoned text, converting the to-be-reasoned text into a corresponding compressed embedded vector sequence based on a preset compressed word list; reconstructing the compressed embedded vector sequence by adopting a preset adapter to generate an embedded vector sequence compatible with the large language model; performing feature extraction on the embedded vector sequence to generate a semantic feature vector; inputting the semantic feature vector into a pre-trained neural network to obtain a candidate word library list to be loaded; and according to the candidate word library list to be loaded, loading the corresponding candidate word library to form a decoding word list, and reasoning based on the decoding word list. According to the method disclosed by the invention, the memory occupation is remarkably reduced while the model performance is ensured, and the efficient deployment and application of the large language model in the end-side equipment are promoted.

The invention relates to the technical field of artificial intelligence, and particularly provides an intelligent writing application method and device based on a large model, and the method comprises the steps: firstly, carrying out the style and specification calibration after a writing module carries out the input analysis and intention understanding, knowledge retrieval and material extraction, and first draft production, and then carrying out the first draft output; the continuous writing module is used for generating new contents which are coherent in semantics, consistent in style and accord with specifications on the basis of the writing module, and then entering the expanding module to deeply and detailedly explain and expand text paragraphs or sentences specified by a user; and finally, the rewriting module adjusts the existing text content, and changes the style, the structure or the expression mode. Compared with the prior art, intellectualization, high efficiency and standardization of writing can be realized, the manuscript quality and processing efficiency are comprehensively improved, the working experience is optimized, and contribution is provided for realizing efficient, simple and intelligent services.

The application discloses an algorithm method for large model long context reasoning, and relates to the technical field of large language models, and comprises the following steps: dividing an input long text sequence into multiple initial text blocks; generating a semantic abstract based on the initial text blocks, and performing clustering analysis on the semantic abstract, merging initial text blocks with similar semantics into semantic super blocks to form a context organization structure with semantic representation; generating key-value cache data of each token in the large model preprocessing stage, grouping the key-value cache data with the semantic super blocks as logical boundaries, and establishing a mapping table recording the storage positions and states of each semantic super block; based on the semantic super blocks, combining the semantic representative vectors of the semantic super blocks, calculating the correlation scores between the query vector and each semantic super block, and learning a block importance prediction model based on context dependency features to provide high-order semantic prior for subsequent attention screening and avoid missing of key information caused by position offset.

The present disclosure discloses a dialogue sentence completion and model training method and device, equipment and a storage medium, relating to the technical field of artificial intelligence, especially to the technical field of natural language processing, human-computer dialogue and the like. The training method of the dialogue sentence completion model comprises: using an encoder to perform encoding processing on a multi-round dialogue sentence sample to generate semantic features, wherein the multi-round dialogue sentence sample comprises a to-be-completed sentence sample; generating a prediction probability value based on the semantic features; using a decoder to perform decoding processing on the semantic features to generate a predicted complete sentence corresponding to the to-be-completed sentence sample; constructing a total loss function based on the prediction probability value and the predicted complete sentence; and adjusting model parameters of at least one of the encoder and the decoder based on the total loss function. The present disclosure can improve the accuracy of the dialogue sentence completion model.

The invention discloses an audit evidence matching method based on a large language model and loop verification, which comprises the following steps of: firstly, acquiring a target audit item and an audit evidence set, inputting the target audit item into the large language model to extract a keyword and generate a semantic feature vector, and performing keyword extraction and feature vector generation on audit evidence; then, text similarity and vector similarity are calculated through a cosine similarity algorithm and an Euclidean distance similarity algorithm respectively, and an initial matching degree is obtained through weighted fusion; and then setting a threshold value to screen to-be-verified evidences, outputting verification suggestions by utilizing a large language model, if the evidences are matched and are to be optimized, adjusting semantic analysis parameters or similarity weights of the model according to rules, and performing repeated cyclic verification until preset conditions are met, thereby generating a final matched evidence list and a to-be-manually rechecked evidence list. Through the multi-dimensional similarity calculation and cyclic verification mechanism, the accuracy of matching of the audit evidence and the target audit item is effectively improved, and the audit working efficiency and quality are improved.

This invention provides an intelligent design method for building walls based on a semantic-physical dual closed loop, comprising the following steps: acquiring project data and environmental element data of the target project, generating semantic tags and project context, retrieving standard provisions based on semantic tags, and compiling the standard provisions into constraint expressions; constructing an executable design grammar for wall design, establishing a bidirectional mapping relationship between the executable design grammar and the building information model, and generating multiple candidate wall schemes conforming to the executable design grammar under the constraints of the constraint expressions; using a surrogate model to perform multiphysics performance evaluation on the candidate wall schemes, triggering a high-precision solver to verify based on the uncertainty evaluation results, and selecting a set of compliant schemes; iteratively optimizing the set of compliant schemes under the penalty mechanisms of hard and soft constraints, outputting a Pareto optimal solution set, parsing the Pareto optimal solution set based on the executable design grammar, and generating design documents and a compliance audit report.

Embodiments of the present application provide a cross-domain multi-round multi-intention recognition method, system, electronic device and storage medium. The method comprises: in a multi-round dialogue, real-time activity voice detection is performed, when dialogue sentence input is detected, an intention recognition module is used to perform intention analysis on the dialogue sentence input in the current dialogue round; it is judged whether the intention of the dialogue sentence in the current dialogue round is in the intention expectation space of the completed dialogue before the current dialogue round, if yes, cross-domain semantic analysis is performed on the dialogue sentence; in the semantic analysis, it is judged whether the dialogue sentence is complete, if not, the dialogue sentence is supplemented to generate a semantic complete sentence; a response sentence is generated for the semantic complete sentence, and is broadcast. Embodiments of the present application introduce the method of intention expectation space to judge whether recognition is needed and whether the interruption needs to be broadcast, break the limitation of the voice recognition model judged from the voice level, and avoid the false interruption caused by noise.

The invention relates to an engineering entity alignment method based on generative semantic enhancement and structural feature reasoning, and belongs to the technical field of knowledge graph construction and multi-source data fusion. Comprising the following steps: constructing a semantic enhanced entity text sequence; embedding the semantic enhanced entity text sequence into a prompt template containing an alignment task instruction, and inputting a large language model to obtain an entity semantic vector containing engineering context logic; utilizing a graph neural network to capture structural topological features among entities, fusing the structural topological features with the entity semantic vectors, calculating preliminary similarity among the entities, and screening out a candidate alignment set; the entity pairs in the candidate alignment set and neighborhood difference information of the entity pairs are converted into a natural language reasoning task, a big language model is used for generating a discriminant reason, and the final alignment probability is calculated. According to the method, a semantic gap caused by sparse engineering data can be effectively broken through, and the accuracy and robustness of cross-professional engineering entity alignment are remarkably improved.

The application relates to a legal text small sample named entity recognition method fusing knowledge prompts and belongs to the field of natural language processing and machine learning. The application firstly defines the mapping relationship of the suggestive template, the label set and the natural word set of the judicial documents, obtains 2760 pieces of ruling and judgment document from the Chinese judicial document network platform; then encodes the embedding vectors of the case statement and the judgment result sentence by adopting a Bert model, obtains the feature representation of the sentence through attention weighting, and generates the category vector of the word in combination with the template guided generative pre-training model GPT; finally, the sentence vector is converted into a span vector by using a full connection layer, a semantic label is generated, and the loss function of the semantic label and the label vector is minimized. The application effectively alleviates the overfitting and inaccurate classification problems caused by the small amount of labeled samples and the different entity type distribution in the legal field, and improves the accuracy of the legal entity recognition and the migration ability of the model.

The application relates to the technical field of text error correction, and discloses an automatic error correction method and system based on a knowledge graph. Based on the lexical semantic association and the syntax dependency rule of the knowledge graph, the surrounding fragments of potential errors of input text are extracted, and semantic similarity is calculated to obtain error association weight data; an influence ranking is generated in combination with the domain importance attribute of the error type, and a to-be-corrected error list is formed after context sentiment tendency analysis adjustment; global semantic representation vectors are constructed based on the knowledge graph, domain categories are recognized, and semantic constraint conditions are generated, a preliminary error correction path is screened, semantic fitness is calculated to determine an optimal path, text correction is completed in combination with a high-priority error correction rule, and a result is output. Through the method, semantic deviation can be effectively avoided, error correction accuracy and pertinence can be improved, the semantic integrity and expression coherence of the corrected text can be ensured, and the problem of insufficient adaptability of traditional error correction methods is solved.

The invention relates to an interactive zero sample composite fault diagnosis method based on fuzzy semantics and FNN, and the method comprises the following steps: S1, collecting vibration signals of different types of faults of a bearing in a rotating machine under different working conditions, and generating a single fault data set and a composite fault data set; s2, constructing a fault diagnosis framework, wherein the fault diagnosis framework comprises a feature extraction module, a semantic construction module, a semantic embedding module, a reasoning module and an interactive expansion module; the feature extraction module extracts single fault sample features, and the semantic construction module generates composite fault generation semantics; the semantic embedding module adopts an FNN to generate prediction semantics; the reasoning module gives a prediction result; s3, training the fault diagnosis framework; and S4, inputting the composite fault data set into the trained fault diagnosis framework, and outputting a fault diagnosis result. According to the method, the self-adaptive capability of the model domain is considered while accurate recognition of the unseen composite fault is ensured, and the method is suitable for composite fault diagnosis of multiple unknown domains.