39 results about "Distributed representation" patented technology

The invention provides an enhanced generation method based on question matching retrieval, and belongs to the field of matching generation, and the method comprises the following steps: S1, a semantic feature coding stage: carrying out real-time feature extraction and vector space mapping on a natural language query input by a user by adopting a deep neural network model, generating high-dimensional distributed representation with semantic representation capability; s2, a knowledge base intelligent retrieval stage: executing multi-dimensional semantic matching in the vectorized knowledge base based on an approximate nearest neighbor search algorithm, and screening out a candidate knowledge set highly related to query semantics through a similarity measurement function; s3, retrieval matching results are automatically associated to the structured knowledge base through the established semantic-knowledge mapping relation, the preprocessed standardized response content is directly obtained, and the response content adopts a multi-modal data organization form and comprises a structured data entity and retains a rich text expression form.

The invention discloses a flood control and emergency rescue plan generation method, which comprises the steps of S1, collecting flood control and emergency rescue original data, performing preprocessing and feature extraction, and labeling and confirming extracted feature data to serve as a training set; s2, performing LoRA fine tuning of a low-rank decomposition matrix on the basis of a pre-trained large language model, and performing large language model training operation through the training set; s3, constructing a flood control and emergency rescue field ontology model, learning distributed representation of the knowledge graph by adopting a graph neural network technology, and dynamically updating the knowledge graph and executing reasoning query according to real-time dangerous case information by establishing a dynamic reasoning mechanism; and S4, constructing a multi-modal information fusion framework which is used for carrying out information fusion processing on the knowledge graph reasoning result and then guiding the big language model to generate an emergency plan as a control signal. According to the method, the technical bottlenecks of low plan generation quality, difficulty in knowledge updating, insufficient multi-source information fusion and the like of a traditional method in a complex dangerous case scene are effectively solved.

The invention discloses an industrial equipment anomaly detection method and system based on word embedding coding and an industrial equipment anomaly detection system based on word embedding coding. Comprising a data acquisition module, a data quality evaluation module, a working condition screening module, a data symbolization processing module, a word embedding coding module, a baseline training module, a real-time abnormal value calculation module and an alarm output module. A distributed representation technology for processing industrial continuous data or time series data is introduced, and word embedding representation of the continuous data is obtained by learning the continuous data by using Transform or word2vec, so that distributed representation using a fixed-width vector and fusing historical data information can be realized; compared with traditional feature engineering directly using original data or based on the original data, the method is more effective; and an auto-encoder is utilized, so that the defects in a traditional anomaly detection method are optimized.

The invention discloses a generation type intelligent large screen construction method and system based on natural language questions and answers, and belongs to the technical field of computer software, and the method comprises the steps: data preprocessing and knowledge base construction: constructing an interface metadata description system based on a data interface, generating a high-dimensional vector through large-scale pre-training, and constructing a knowledge base; dense distributed representation of text semantics is realized; performing question segmentation and extraction: performing automatic segmentation on an intelligent large screen generation question proposed by a user by adopting a generative pre-training model, and matching an interface corresponding to an optimal interface description based on a model reasoning capability; index data acquisition: performing iterative processing by adopting a polling mode, and acquiring index data according to the interface metadata information; and large screen code generation and analysis to realize code generation of an intelligent large screen. According to the method, the problems that traditional large screen development needs to be participated by professional technicians and the response period is long are solved, automatic analysis of multi-source heterogeneous data and visual code generation are achieved, and finally an intelligent large screen application capable of being directly deployed is generated.

This system or method provides a reading comprehension support system that allows natural language input as query text and presents the reader with sections of the text that are highly relevant to the input text. [Solution] The reading support system includes a document reading unit 101 that reads the target document, a block division unit 103 that divides the target document into multiple blocks, a distributed representation acquisition unit 104a that acquires word distributed representations for each of the multiple blocks, a question input unit 102 that reads the question, a distributed representation acquisition unit 104b that extracts words contained in the question and acquires word distributed representations, and a similarity calculation unit 106 that compares the word distributed representations in the question and each of the multiple blocks and calculates the similarity. The similarity calculation unit searches for words that match words contained in the question from among the words contained in the block, and for the matching words, it calculates the similarity between the word distributed representation in the block and the word distributed representation in the question.

An information processing apparatus comprises: an acquisitor to obtain document information, including character strings and position data from document image data; a converter to transform the acquired document information into a distributed representation; an information extractor to identify a character string corresponding to an item specified by a prompt, using a large language model by inputting the prompt with the acquired document information; a storage unit to save the document information, distributed representation, and extraction results, associating them with the document; and a selector to choose a reference document from previously processed documents based on distributed representations. The prompt for processing a new document includes details about the selected reference document.

The application belongs to the field of subcellular localization prediction, and provides an lncRNA subcellular localization prediction method and system for unbalanced data, in order to fully utilize the sequence information of lncRNA, physical and chemical mode features and distributed representation features of nucleic acids are extracted, and two basic classifiers including a convolutional neural network and a gated recurrent unit are integrated, a subcellular classification result is obtained through adaptive weighted averaging of the two models, and subcellular localization is carried out based on the classification result, so that the precision of subcellular localization prediction on an unbalanced data set can be improved. In order to solve the problem that the model performs poorly on the unbalanced data set, a label distribution margin perception loss function is used in the training process. Compared with traditional machine learning models and existing predictors, the method has stronger tolerance to class imbalance and robustness.

To provide a scheme that generates a search keyword using a generative AI.SOLUTION: An information processing system converts each character string constituting a first data group to be a search target into a distributed representation using a distributed representation model with the character string as input and the distributed representation as output, that has been learned using the first data group or a second data group different from the first data group as learning data, inputs a condition input by a user into a generative AI to obtain a keyword related to the condition as a generative AI keyword, obtains the distributed representation corresponding to the generative AI keyword using the distributed representation model, obtains, as a similar keyword, a character string from among the character strings constituting the first data group whose distributed representation has a similarity degree of a predetermined value or more compared to the distributed representation of the generative AI keyword, and searches the first data group using the generative AI keyword and the similar keyword to extract a corresponding search target.SELECTED DRAWING: Figure 7

The invention discloses a decision-making method and system for an organic semiconductor luminescent material, a terminal and a medium, and the method comprises the steps: integrating multi-source heterogeneous data in the field of organic semiconductor luminescent materials, and constructing a high-quality heterogeneous knowledge base; the method comprises the following steps of: converting natural language query of a user into structured query by adopting a semantic analysis model adaptive to BERT based on a material field, and dynamically constructing a knowledge sub-graph related to a task; constructing a symbol reasoning engine based on domain rules, executing interpretable logical reasoning to obtain a symbol reasoning conclusion, adopting a material domain heterogeneous graph representation learning model to perform distributed representation learning and multi-task optimization to obtain a learning reasoning conclusion, integrating the symbol reasoning conclusion and the learning reasoning conclusion, and generating a final decision suggestion. According to the method, the efficiency and the accuracy of material research and development are improved by constructing the multi-source heterogeneous knowledge base, designing the self-adaptive scheduling engine and realizing multi-level reasoning fusion.

The application discloses a composition automatic scoring method based on multi-stage learning, and the method comprises the following steps: S1, extracting the shallow language features, emotional features and theme relevance features of the composition; S2, theme relevance feature extraction; S3, construction of a beautiful sentence identification model and extraction of composition style features; S4, training of a base learning machine; and S5, composition vector distributed representation and feature fusion model training and prediction. The application is applied to the field of automatic composition scoring, and a comprehensive and multi-dimensional composition scoring feature is designed for Chinese composition scoring, the detection and discovery of beautiful sentences in the composition are realized, and the beauty degree of language expression in the composition is better considered; meanwhile, the composition automatic scoring based on multi-stage learning is proposed, and multi-angle composition features are effectively combined for composition scoring.

A reference map generator clusters, into a semantic map, a set of data documents selected according to at least one criterion and associated with a medical diagnosis. A parser generates an enumeration of measurements occurring in the set of data documents. A representation generator generates for each measurement in the enumeration, a sparse distributed representation (SDR). The method includes storing, by a processor on a second computing device, in each of a plurality of memory cells on the second computing device, one of the generated SDRs. A diagnosis support module receives a document comprising a plurality of measurements. The representation generator generates a compound SDR for the document. Each of the plurality of bitwise comparison circuits determine a level of overlap between the compound SDR and the stored generated SDR. The diagnosis support module provides an identification of the medical diagnosis associated with a stored SDR.

To improve the accuracy of an answer obtained by a large-scale language model by preventing a sentence piece extracted from an inappropriate document from being mixed with a question.SOLUTION: The database unit stores, as one record, a sentence piece extracted from each of a plurality of documents, a sentence piece feature vector representing a distributed representation of the sentence piece, and document identification information for identifying the document from which the sentence piece is extracted. The similarity calculation unit calculates a similarity with the question sentence for each sentence piece by a question sentence feature vector representing a distributed representation of the input question sentence and a sentence piece feature vector of each record stored in the database unit. A document specification part specifies a document related to the question sentence from among the plurality of documents on the basis of the similarity calculated for each sentence piece. The sentence piece extraction unit extracts at least one sentence piece from the specified document in descending order of similarity. The answer generation unit generates an answer to the question sentence on the basis of the question sentence and the extracted sentence piece.SELECTED DRAWING: Figure 7

The application discloses a decision-making method, system, terminal and medium of an organic semiconductor light-emitting material, and the method comprises the following steps: integrating multi-source heterogeneous data in the field of organic semiconductor light-emitting materials, and constructing a high-quality heterogeneous knowledge base; adopting a semantic analysis model based on a material field adaptive BERT to convert a user natural language query into a structured query, and dynamically constructing a knowledge subgraph related to a task; constructing a symbolic reasoning engine based on a field rule, performing an interpretable logical reasoning, obtaining a symbolic reasoning conclusion, adopting a material field heterogeneous graph representation learning model to perform distributed representation learning and multi-task optimization, obtaining a learning reasoning conclusion, integrating the symbolic reasoning conclusion and the learning reasoning conclusion, and generating a final decision-making suggestion. Through the construction of a multi-source heterogeneous knowledge base, the design of an adaptive scheduling engine and the realization of multi-level reasoning fusion, the efficiency and accuracy of material research and development are improved.

An information processing apparatus extracts a phrase corresponding to a named entity included in an input sentence used for training a natural language processing model from data in which a phrase and a classification name of a named entity are associated with each other, adds the extracted phrase to the classification name corresponding to the extracted phrase among a plurality of different classification names which are set in advance, derives a similarity between a distributed representation of each word included in the input sentence and a distributed representation of each of the plurality of classification names to which the phrase is added, and select a classification name of each word included in the input sentence from among the plurality of classification names based on the derived similarity.

This invention relates to a named entity recognition method based on a pre-trained language model and a progressive convolutional network, comprising the following sequential steps: encoding natural language based on the pre-trained language model to obtain a representation set LS; inputting the representation set LS into a progressive convolutional network module, and using the progressive convolutional network module to progressively fuse the encodings of adjacent layers from low to high levels to obtain an aggregated distributed representation AR that integrates the features of all layers of the pre-trained language model. c ;Utilizing the CRF model, i.e., Conditional Random Field, to decode and aggregate distributed representations (AR) c This invention achieves named entity recognition. Instead of introducing external knowledge or operations to enhance entity information and improve named entity recognition accuracy, it focuses on the results obtained using the available computing power. It utilizes the proposed progressive convolutional network module to extract the full-layer representation of the pre-trained language model, overcoming the deficiency of insufficient information mining from the pre-trained language model and reducing the complexity of introducing external data for computation.

In question answering systems, the accuracy of answers obtained using large-scale language models can be improved. [Solution] The question answering device creates a virtual question sentence from a text fragment and stores the text fragment and the virtual question sentence vector, which represents the virtual question sentence created from the text fragment in distributed representation, as one record in the database. For each record stored in the database, the question answering device calculates the similarity between the virtual question sentence vector contained in that record and the question sentence vector, which represents the input question sentence in distributed representation. Based on the similarity calculated for each record, the question answering device extracts a predetermined number of text fragments from the database, which are contained in a predetermined number of records. The question answering device generates an answer to the question sentence based on the question sentence and the predetermined number of extracted text fragments.

A method encodes compositional data structures by receiving component sparse distributed representation arrays having a predetermined array length and target sparsity level, applying position-specific permutation transformations to encode ordinal position information, combining position-encoded arrays through bitwise union to generate an intermediate array, and processing the intermediate array through a dual-phase sparsity reduction procedure comprising a coarse additive phase and a fine subtractive phase controlled by a sparsity overshoot threshold to generate a composite encoded array. The dual-phase procedure converges in substantially constant iterations for 4 or more components with final sparsity tightly controlled around the target. A decoding method applies inverse position-specific transformations to generate position-decoded arrays, computes overlap scores with candidate components through bit counting operations, and determines component identities based on threshold comparison. Scalable decoding may use triadic associative memory. Applications include searchable compression, privacy-preserving analytics, and efficient neural network embeddings.

The invention provides an information processing apparatus, an information processing method, a computer program product, and a recording medium. The information processing device includes: an acquisition unit that acquires document information included in image data of a document; a conversion means for converting the acquired document information into a dispersed representation; an information extraction unit that inputs a cue including the acquired document information into the large-scale language model, performs reasoning based on the large-scale language model, and extracts a character string corresponding to an item indicated by the cue; a storage unit that associates and stores in a storage unit document information relating to the document, the dispersion performance, and the extraction result by the information extraction unit; and a selection means for selecting a reference document from the documents that have been processed in the past on the basis of the distributed representation relating to the document that has been processed in the past and the distributed representation relating to the documents that have been processed in the past and stored in the storage unit, the presentation input to the document that has been processed including information relating to the selected reference document.

This invention discloses a bio-inspired text sequence processing method, belonging to the field of text sequence processing, and applied to sequence retrieval and sequence recovery tasks. This method mimics the micropillar structure of the human cerebral cortex, encapsulating a large number of parallel Spiking neurons within each micropillar structure; it incorporates synaptic delay and Theta oscillation mechanisms to ensure periodic learning and prediction of text sequences; it designs a sparse temporal group coding scheme to transform input text characters into sparse distributed representations; and it proposes Spiking-based unsupervised learning rules to realize the storage and association process of text sequences. By periodically storing and associating the distributed representations of input sequence characters, this method can complete sequence retrieval tasks from partial context and sequence recovery tasks from damaged information, providing a new approach to the construction of artificial association systems and expanding the application scope of Spiking-based neuromorphic chips.

The invention discloses an emergency resource classification method and system based on natural language processing, and relates to the technical field of data classification. Aiming at the problems of low efficiency and poor accuracy of the existing emergency resource classification method, the adopted scheme comprises the following steps: S1, carrying out data cleaning, word segmentation and part-of-speech tagging on input emergency resource text data; s2, extracting text word frequency features, weighting text words, obtaining word distributed representations, and combining the word distributed representations into text vectors; s3, constructing an emergency resource classification model, and introducing an attention mechanism into an input layer or a middle layer of the model; carrying out model training and optimization by using the labeled emergency resource text data, and enabling the model to output probability distribution of emergency resource texts belonging to different categories; and S4, obtaining a new emergency resource text, sequentially executing the previous steps, outputting probability distribution that the text belongs to different categories by the model, and selecting the category with the highest probability as a classification result. According to the invention, the emergency resources can be automatically classified and managed.

The application discloses a kind of object group determination method, device, equipment, storage medium and program product, involve artificial intelligence and data processing technical field.The method comprises: obtaining the aggregation zone structure graph corresponding to reference object group, and the multiple aggregation zone structure graphs corresponding to the network to be detected;Obtain the distributed representation corresponding to reference object group, and the multiple distributed representations corresponding to the network to be detected, the distributed representation is used to represent the structural features of aggregation zone structure graph;According to the distributed representation corresponding to reference object group, determine the target aggregation zone structure graph from the multiple aggregation zone structure graphs corresponding to the network to be detected;Obtain the target object group corresponding to target aggregation zone structure graph.The application determines the object group by supporting the aggregation zone level, without determining object one by one, and considering the overall correlation of object group, so as to improve the determination efficiency and determination accuracy of object group.

A speech recognition device according to the present disclosure comprises: a speech distributed representation sequence conversion unit that generates an intermediate acoustic feature sequence through conversion by means of a first neural network; a symbol distributed representation sequence conversion unit that generates an intermediate character feature sequence through conversion by means of a second neural network; a blank estimation unit that accepts the intermediate acoustic feature sequence and the intermediate character feature sequence as input, and outputs a first output probability distribution by means of a third neural network; a label estimation unit that accepts the intermediate acoustic feature sequence and the intermediate character feature sequence as input, and outputs a second output probability distribution by means of a fourth neural network; an HAT loss calculation unit that accepts a third output probability distribution, which is obtained by integration of the first output probability distribution and the second output probability distribution, and a ground-truth symbol sequence as input, and calculates a first loss for HAT; and an internal acoustic model loss calculation unit that accepts the third output probability distribution and the ground-truth symbol sequence as input, and uses the same model parameters as the HAT loss calculation unit.

The present invention improves analysis accuracy in analyzing a document that includes text information describing a substance. A processing method according to the present invention includes: a first output step for outputting a first distributed representation based on structure information pertaining to a substance having a substance name included in a document; a second output step for outputting a second distributed representation based on text information describing a substance having a substance name included in the document; and a training step for training a model by using the first distributed representation and the second distributed representation.

The application provides an entity relation joint extraction method and system based on a triple forest, which comprises the following steps: obtaining corpus to be subjected to entity relation extraction, and obtaining a sentence and a corresponding word sequence thereof; inputting the word sequence into a BERT model, performing word segmentation on the word sequence by the BERT model to obtain a subword sequence, encoding the subword sequence by using the BERT model to obtain a distributed representation of the sentence; inputting the distributed representation into a CRF model to label entities in the sentence and obtain vector representations of the entities; inputting the entity vectors, obtaining hidden layer vectors of interaction information between entities contained in the entity vectors and interaction information between the entities and the input sentence by a multi-head attention mechanism in a TransformerDecoder module; taking the hidden layer vectors as initial states and initial hidden layer units of a Tree-RNN, inputting the entity representations into the Tree-RNN, generating relations participated in by a head entity of a root node of the Tree-RNN, selecting tail entities of the head entity according to the head entity and the corresponding relations, thereby generating an overlapping triple tree, and further decoding to obtain entity relation triplets.

The invention discloses a multilayer consensus knowledge base construction method based on distributed representation, and particularly relates to the field of semantic analysis, which comprises the following steps: receiving text data, executing statement segmentation and keyword extraction, and generating a knowledge node set; marking a timestamp and a context reference relationship for each knowledge node, and generating a corresponding distributed vector representation in a unified embedding space; and when new text data is received, generating new knowledge nodes based on the original historical knowledge node set. A reverse influence calculation and redefinition layer generation mechanism is introduced after knowledge nodes are generated, so that posterior knowledge can perform reverse correction and redefinition on historical nodes in a distributed semantic mode, and bidirectional evolution and semantic self-consistency of a knowledge system in a time dimension are realized; the problems that in the prior art, time error-sequence emerging cannot be processed, and future knowledge is lacked to remodel a past semantic mechanism are solved.

This enables the presentation of the level of abstraction of the words used to represent skills when creating a hierarchical skill list. [Solution] The skill word acquisition unit 2 acquires multiple skill words selected by the skill list creator from a list of multiple skill words. The skill description acquisition unit 3 acquires multiple skill descriptions generated using the generation AI 10. The distributed representation acquisition unit 4 acquires a distributed representation for each skill word, converted from each of the multiple skill descriptions using the embedding model 20. The abstraction level (standard deviation) calculation unit 5 calculates the standard deviation of the distributed representation in at least one dimension of the distributed representation (vector) of each of the multiple skill descriptions for each skill word. The standard deviation (magnitude) of at least one dimension corresponds to the abstraction level (height) of that skill word.

The invention discloses a steel identification intelligent identification method and system based on deep learning. The method comprises the following steps: carrying out image acquisition and preprocessing to form a preprocessed steel surface image sequence; constructing a hierarchical time sequence memory network and completing structure initialization; positioning a steel identification area, and generating a steel identification area image sequence; extracting steel identification visual structure features and encoding to generate a sparse distributed representation sequence; performing time stability judgment and outputting a stable time interval index set; generating a character sequence prediction state sequence according to the stable time interval index set; executing time consistency judgment to determine a steel identification recognition result; performing combination legality verification, and generating coding rule consistency constraint information; information is fed back to the memory updating process, and a steel identification recognition result subjected to coding rule consistency constraint is output. Layered time sequence memory deep learning is utilized, steel material identification time sequence consistent identification is achieved, and the method is high in stability and reliability.

To provide an information processor capable of accurately extracting a character string corresponding to a desired item from image data of a document.SOLUTION: The information processing apparatus includes acquisition means for acquiring document information (character string and position information) included in image data of a document, conversion means for converting the acquired document information into a distributed representation, extraction means for extracting a character string corresponding to an item instructed by the acquired document information by inputting a prompt including the acquired document information to a large-scale language model and performing inference by the large-scale language model, storage means for storing the document information related to the document and the extraction result by the information extraction means in a storage unit in association with each other, and selection means for selecting a reference document from previously processed documents based on the distributed representation related to the document to be processed and the distributed representation related to the previously processed document stored in the storage unit.SELECTED DRAWING: Figure 2

The present invention provides a method and a terminal device for recommending treatment measures for unfavorable structural planes of arch dam abutments based on a knowledge graph, comprising the following steps: constructing an ontology model of the knowledge graph of geological information of unfavorable structural planes of arch dam abutments, extracting geological entity information related to the unfavorable structural planes, and improving the data layer of geological information; constructing an ontology model of the knowledge graph of treatment measures for unfavorable structural planes of arch dam abutments, extracting treatment measure entity information related to the unfavorable structural planes, and improving the data layer of treatment measures; establishing a corresponding relationship between the geological information of the unfavorable structural planes and the treatment measures to form a knowledge graph model of treatment measures for unfavorable structural planes of arch dam abutments; carrying out knowledge graph completion based on an inference method of distributed representation learning; and inputting the geological survey information of the unfavorable structural planes to be processed into the graph to recommend corresponding treatment measures. The present invention performs structured knowledge expression on the geological survey information and the treatment measures for unfavorable structural planes of arch dam abutments, and can quickly adjust the treatment measures for unfavorable structural planes as the geological information is continuously improved.

The invention discloses a non-coding RNA-disease association prediction method, device and equipment and a storage medium, and is applied to the technical field of biological information, and the method comprises the steps: constructing a heterogeneous biological network, employing a graph neural network to learn the distributed representation of a biological entity from a network topology structure and an entity attribute feature based on the heterogeneous biological network, and obtaining the distributed representation of the biological entity; different view information is fused through an attention mechanism; establishing a multi-task learning model, and extracting a shared biological mode suitable for all prediction tasks from the graph neural network; generating prediction results of different association types through a decoder based on the shared biological mode; according to the method, multi-task training is optimized through reinforcement learning dynamic parameter adjustment, heterogeneous and attribute information is fused through a double-view graph network, a multi-task framework is unified to share a biological mode, and special decoding is carried out, so that collaborative improvement of precision and generalization ability in ncRNAs-disease association prediction is realized.