Patents
Literature
Hiro is an intelligent assistant for R&D personnel, combined with Patent DNA, to facilitate innovative research.
Hiro

90 results about "Semantic enhancement" patented technology

Apparatus and methods for developing conversational applications

Apparatus with accompanying subsystems and methods for developing conversational computer applications. As a user interface, the apparatus allows for a user to initiate the conversation. The apparatus also answers simple and complex questions, understands complex requests, pursues the user for further information when the request is incomplete, and in general provides customer support with a human like conversation while, at the same time, it is capable to interact with a company's proprietary database. As a development tool, the apparatus allows a software developer to implement a conversational system much faster than takes, with current commercial systems to implement basic dialog flows. The apparatus contains three major subsystems: a state transition inference engine, a heuristic answer engine and a parser generator with semantic augmentations. A main process broker controls the flow and the interaction between the different subsystems. The state transition inference engine handles requests that require processing a transaction or retrieving exact information. The heuristic answer engine answers questions that do not require exact answers but enough information to fulfill the user's request. The parser generator processes the user's natural language request, that is, it processes the syntactical structure of the natural language requests and it builds a conceptual structure of the request. After the parser generator processes the user's request, a main process broker feeds the conceptual structure to either the heuristic answer engine or to the state transition inference engine. The interaction between the main process broker and the subsystems creates a conversational environment between the user and the apparatus, while the apparatus uses information from proprietary databases to provide information, or process information, during the course of the conversation. The apparatus is equipped with a programming interface that allows implementers to declare and specify transactions based requests and answers to a multiplicity of questions. The apparatus may be used with a speech recognition interface, in which case, the apparatus improves the recognition results through the context implicitly created by the apparatus.
Owner:GYRUS LOGIC INC

Disease specific ontology-guided rule engine and machine learning for enhanced critical care decision support

A disease-specific ontology crafted by a consensus of expert clinicians may be used to semantically characterize / provide semantic meaning to dynamically changing patient electronic medical record (EMR) data in critical care settings. Hierarchical, directed node-edge-node graphs (concept maps or Vmaps) developed with an end-user friendly graphical user interface and ontology editor, can be used to represent structured clinical reasoning and serve as the first step in disease-specific ontology building. Disease domain Vmaps reflecting expert clinical reasoning associated with management of acute illnesses encountered in critical care settings (e.g. ICUs) that extend core clinical ontologies, developed and reviewed by experts, are in turn extended with existing medical ontologies and automatically translated to a domain ontology processing engine. Semantically-enhanced EMR data derived from the ontology processing engine is incorporated into both real-time ‘track and trigger” rule engines and machine learning training algorithms using aggregated data. The resulting rule engines and machine-learnt models provide enhanced diagnostic and prognostic information respectively, to assist in clinical dual modes of reasoning (analytical rules and models based on experiential data) to assist in decisions associated with the specific disease in acute critical care settings.
Owner:COMP TECH ASSOC INC

A human posture regression method and system based on point cloud semantic enhancement

A method and system for human posture regression based on semantic enhancement of point cloud is provided in that embodiment of the present invention. The method comprises steps extracting the point cloud features of the hand point cloud data, and classifying them point by point, obtaining semantic segmentation information of hand point cloud data, Hand-point cloud data being semantically enhancebased on semantic segmentation information, based on the hand point cloud data after semantic enhancement, and geometrically transforming the hand posture prediction result, so that that regression result of the hand posture are obtained. The method of geometric transformation of input data and output data by network learning makes the method of human pose estimation more robust to the geometric transformation of input data. The semantic information of input point cloud classification sub-network and attitude regression sub-network are fused effectively, and the performance of human pose estimation is further improved.
Owner:TSINGHUA UNIV

Semantic mapping and positioning method based on priori laser point cloud and depth map fusion

The invention discloses a semantic mapping and positioning method based on priori laser point cloud and depth map fusion. The method comprises the steps of S1, collecting priori laser point cloud data; S2, acquiring a depth image and an RGB image, generating RGB-D point cloud based on the depth image, and initializing and registering priori laser point cloud and RGB-D point cloud; S3, camera poseconstraints are provided by the registered prior laser point cloud to perform camera pose correction; S4, a three-dimensional geometric point cloud map is created by adopting a front and back window optimization method; S5, geometric increment segmentation is carried out on the three-dimensional geometric point cloud map, object recognition and semantic segmentation are carried out on the RGB image, geometric increment segmentation and semantic segmentation results are fused, and a 3D geometric segmentation map of semantic enhanced geometric segmentation is obtained; and S6, semantic association and segmentation probability allocation updating is carried out on the object to complete construction of a semantic map. Accumulated errors of large-scale indoor mapping and positioning can be effectively eliminated, and precision and real-time performance are high.
Owner:AEROSPACE INFORMATION RES INST CAS

Scene character recognition method and system based on semantic enhancement encoder decoder framework

PendingCN111753827AEffective bluEffective situationNeural architecturesCharacter recognitionEncoder decoderGlobal information
The invention provides a scene character recognition method and system based on a semantic enhancement encoder decoder framework, and the method comprises the steps of correcting a text in any shape on a target image into a horizontal text, and obtaining a corrected image; inputting the corrected image into a convolutional neural network to extract visual features, and extracting sequence information from the visual features by using a recurrent neural network; predicting global semantic information according to the sequence information; and initializing the state of a GRU (Gateway Recirculation Unit) based on an attention mechanism by utilizing the global semantic information, calculating an attention weight according to the visual feature and the hidden state of each decoding time of theGRU, weighting the visual feature according to the attention weight, and predicting each character on the image. According to the invention, global information can be effectively utilized to overcomethe defect that local information is used in an existing method, and meanwhile, a gap between visual information and semantic information is reduced, so that the model can better process a low-quality image.
Owner:INST OF INFORMATION ENG CAS

Deep learning sentiment analysis model based on semantic enhancement and analysis method thereof

InactiveCN110502753AEnhancing the ability to capture emotional semanticsFix workNeural architecturesSpecial data processing applicationsAnalytic modelFeature vector
The invention discloses a deep learning sentiment analysis model based on semantic enhancement, and the model consists of six layers: a word embedding layer, a sentiment semantic enhancement layer, aCNN convolution sampling layer, a pooling layer, an LSTM layer, and a sentiment classification layer in sequence from the bottom to the top. The word embedding layer converts words of the sentences into low-dimension word vectors; the emotion semantic enhancement layer is used for enhancing emotion semantics of the model; the CNN convolution sampling layer is used for automatically extracting wordfeatures; the pooling layer is used for reducing the dimension of the feature vector; the LSTM layer is used for capturing a long-distance dependency relationship in the statement and memorizing long-time dependency serialized information; and the sentiment classification layer adopts Softmax to perform sentiment classification. According to the method, the LSTM layer is added, so that the emotion analysis accuracy can be improved, and meanwhile, the emotion semantic enhancement layer is added, so that the emotion semantics of the model is enhanced, and the emotion analysis effect is improved; the invention further discloses a sentiment analysis method based on the deep learning sentiment analysis model, and the accuracy of Chinese short text sentiment analysis can be improved.
Owner:KUNMING UNIV OF SCI & TECH

Short text classification method based on semantic enhancement

The invention discloses a short text classification method based on semantic enhancement. The method comprises the steps of 1, constructing a short text classifier, obtaining a short text training setrelevant to the field from internet resources, expanding corpuses and training word vectors for each short text, and training the short text classifier; 2, after expanding the corpuses and training the word vectors for each to-be-classified short text, inputting the to-be-classified short texts to the short text classifier obtained in step 1 for classification to obtain a classification result. By means of the short text classification method based on the semantic enhancement, the semantics of the short texts is enhanced and the texts are classified, aiming at the features of the short textsthat the information amount is small and the semantics is sparse, a method of expanding the corpuses with high quality and training the word vectors with high precision is utilized to conduct semanticenhancement representation on the short texts; meanwhile, an efficient text classification algorithm is utilized, the finite features of the texts are captured to the greatest extent, and the training time of the classifier is effectively shortened.
Owner:中国人民解放军军事科学院军事科学信息研究中心

Method and system for enhancing file entity association degree based on knowledge graph

The invention discloses a method and a system for enhancing file entity association degree based on a knowledge graph. The method comprises the following steps: obtaining archive text data; identifying the archive text data by utilizing an entity identification model, and generating instance data of a defined entity; identifying the instance data of the defined entity by using a relationship extraction model, and generating a minimum unit in a knowledge graph; using a knowledge fusion model to carry out deduplication preprocessing on the minimum unit in a knowledge graph, establishing partition index sub-documents, searching for matched entities according to text similarity or structural similarity, and performing knowledge fusion through a preset entity alignment algorithm to enhance theassociation degree of archive entities. The main functions of intelligent file collection and filing, data processing and analysis and file resource semantic enhancement are achieved through entity recognition, relation extraction and fusion technologies, powerful support is provided for semantic association and intelligent development of file management, and the file data association degree and the file data utilization rate are increased.
Owner:AGRI INFORMATION INST OF CAS

Knowledge graph construction method for intelligent power grid customer service questions and answers

The invention discloses a knowledge graph construction method for intelligent power grid customer service questions and answers, and the method comprises the following steps: S1, carrying out the entity and assertion recognition of a question through an LSTM model, and outputting the entity and assertion of the question; S2, performing semantic enhancement processing on the entity and the assertion to obtain a matched enhanced text, and constructing a local knowledge graph; S3, based on the matched enhanced text, performing answer query to obtain a final answer of the question; and S4, constructing an integer linear programming model according to the satisfaction of the user to the final answer and the maintenance cost of the knowledge graph, and updating the local knowledge graph. According to the method, high entity and assertion recognition accuracy can be achieved, power grid domain knowledge is automatically mapped into the local knowledge graph, online updating of the service knowledge base is rapidly achieved, and high-accuracy answers are achieved with low response delay.
Owner:STATE GRID CORP OF CHINA +1

Zero sample classification method and device based on semantic enhancement of encyclopedic knowledge

The invention relates to the fields of pattern recognition, machine learning and computer vision, and provides a zero sample classification method based on semantic enhancement of encyclopedic knowledge, which aims to solve the problem that the existing zero sample image classification method cannot give consideration to both word vector language information range and processing efficiency. The zero sample classification method comprises the steps of: S1, classifying unknown-class images by means of a trained convolutional neural network classifier, and subjecting semantic features of classification result tags to convex combination to serve as semantic features of the unknown-class images; S2, and classifying the semantic features of the unknown-class images obtained in the step S1 and semantic features in a pre-constructed semantic feature space by means of a nearest neighbor classifier, so as to obtain final classes of the unknown-class images. The zero sample classification method and a zero sample classification device provided by the invention enhance the global information of word vectors, so as to improve the accuracy of image zero sample classification.
Owner:INST OF AUTOMATION CHINESE ACAD OF SCI

Document-level event argument extraction method based on sequence labeling

The invention requests to protect a document-level event argument extraction method based on sequence labeling, which comprises the following steps of obtaining Wikipedia priori knowledge related to a corpus entity, and generating a word span entity semantic enhancement embedding representation; splicing the word span entity semantic enhancement embedded representation with a context representation obtained by a pre-training language model to obtain word vector input of an embedded layer; inputting the word representation into a multi-span bidirectional recurrent neural network to obtain a multi-span context feature representation of the word; inputting the multi-span context feature representation into a context attention mechanism module and a gated attention mechanism module, and obtaining a context semantic fusion feature representation of the word; and finally, carrying out event argument extraction on the output feature representation by adopting sequence labeling, and carrying out event argument extraction on an unknown document by utilizing an optimal model obtained by training. According to the method, the extraction effect of the document-level event argument is effectively improved by integrating priori knowledge and multi-span upper and lower semantic feature representation.
Owner:CHONGQING UNIV OF POSTS & TELECOMM

Target detection method based on information enhancement

The invention discloses a target detection method based on information enhancement, and aims to overcome the defect of low precision of a single-stage detection method. According to the technical scheme, a target detection system composed of a feature extraction module, a semantic improvement module, a feature selection module, a feature fusion module and a detection module is constructed; a target detection network is trained by adopting the training data set; feature extraction, semantic improvement, feature selection and feature fusion are carried out on a single-frame image by adopting thetrained target detection system; and the position and category of the target are identified. The semantic enhancement module enriches semantic information of multi-scale features; the feature selection module enhances useful information in different scale feature maps by adopting an attention module, useless information is inhibited, and the purpose of information enhancement is achieved; the feature fusion module fuses the global semantic feature map subjected to feature selection to the multi-scale feature map, so that each feature map has more accurate position and semantic information, and the detection precision is improved.
Owner:NAT UNIV OF DEFENSE TECH

Official document abstract generation model combining extraction type and generation type

ActiveCN110119444AEnhancing Semantic MeaningSolve the problem of missing training dataSemantic analysisEnergy efficient computingLabeled dataData mining
The invention provides an official document abstract generation model combining an extraction type and a generation type. The extraction type abstract and the generation type abstract are combined andofficial document data is screened and pre-processed. Meanwhile, the semantic meaning of weak label data generated by the extraction type abstract is enhanced. An official document text abstract automatic generation model is learned to realize automatic generation of the official document abstract. Compared with a traditional abstract generation method based on end-to-end and adding an attentionmechanism, the method has the advantages that the problem of lack of training data is solved, data screening and semantic enhancement are carried out on the basis of the characteristics of the official document data, and the semantic meaning of the official document text can be represented more accurately.
Owner:CETC BIGDATA RES INST CO LTD

A text semantic analysis method and device

ActiveCN109740158ARich semantic representationImprove calculation convergence speedCharacter and pattern recognitionSpecial data processing applicationsSemantic enhancementTarget text
The invention discloses a text semantic analysis method and device, and the method comprises the steps: obtaining the vector representation of a given text, and generating a coding vector of the giventext according to the vector representation; wherein the given text comprises a first text and a second text; wherein the coding vectors comprise a first coding vector and a second coding vector; generating a first attention and a second attention according to the first coding vector and the second coding vector, and generating a global information vector according to the first attention and thesecond attention; obtaining a semantic enhancement vector from the global information vector; and analyzing a target text corresponding to the second text in the first text according to the semantic enhancement vector. Compared with the prior art, the technical scheme of the invention has the advantages that the semantic enhancement vector is acquired from the global information vector, and the target text corresponding to the second text in the first text is analyzed according to the semantic enhancement vector, so that the semantic representation of the vector on the text is enriched, the calculation convergence speed is increased, and the accuracy and the efficiency are improved.
Owner:安徽省泰岳祥升软件有限公司

Semantic enhanced scene text recognition method and device

The invention provides a semantic enhanced scene text recognition method and device, and the method comprises the steps of extracting a visual feature map and a context feature sequence of a scene text image through an encoder of a scene text recognition model; and determining enhanced feature expression based on the visual feature map, the context feature sequence and the position code of the feature map; obtaining global visual information and semantic information of a scene text image; adopting a specially designed recurrent neural network unit for decoding by a decoder, wherein the unit can balance independence and correlation of context information; and performing multi-head attention operation on the implicit state vector and the expanded enhanced feature expression to obtain a local apparent feature vector. The local apparent feature vector and the hidden layer output of the recurrent neural network unit jointly participate in character prediction at the current moment, so that the correlation between semantic information and visual information is enhanced. The multi-head attention mechanism design can capture saliency information and auxiliary information of the features, so that the accuracy of a scene text recognition result is relatively high.
Owner:INST OF AUTOMATION CHINESE ACAD OF SCI

Semantic enhanced hash method for zero-sample image retrieval

A semantic enhanced hash method for zero-sample image retrieval belongs to the technical field of computers and comprises the following steps: 1) performing image feature semantic alignment; 2) maintaining a domain structure; 3) hash code learning; 4) constructing and optimizing a total objective function; and 5) hash function learning for new data. The method mainly aims at solving the problem oflarge-scale image retrieval, and due to the fact that large-scale image data is generated from the Internet, for some newly-generated affairs and new categories, it is difficult for an existing algorithm to collect enough training pictures of new affairs to train a retrieval model. Therefore, the category semantic space is used as the middle transition space between the image visual features andthe binary codes, alignment of the visual space and the category semantic space is achieved, and the purpose of migrating knowledge from visible data to invisible data is achieved. Experimental verification shows that knowledge can be effectively learned from visible data and migrated to an invisible class, and the problem of zero-sample image retrieval is solved.
Owner:DALIAN UNIV OF TECH

Ontological learning method applicable to Web service description

The invention discloses an ontological learning method applicable to Web service description. The method comprises the following steps of, firstly, collecting Web service description documents, obtaining input and output parameters in the documents and preprocessing every input and output parameter; secondly, utilizing an hHDP (h heuristic dynamic programming) algorithm to generate levels of a subject through a top-bottom learning method; thirdly, utilizing a sampling method of 'Chinese restaurant problem' to estimate the subject of every level; lastly, obtaining a representative word to construct an initial ontological body and enhancing the semantics of the ontological body through semantic enhancing rules to form a final ontological body. The ontological learning method applicable to Web service description has the advantages of having good universality by generating the ontological body for Web service unsupervised learning with WSDL (web services description language) description; being capable of finding out implicit semantic level relationship and meanwhile guaranteeing the richness of the learned ontological semantics through the semantic enhancing rules, thereby having good ontological learning effects; being capable of supporting Web service discovery and recommendation and having wide applicability due to the fact that the learning ontological body can serve for Web service semantic annotation.
Owner:WUHAN UNIV

Emotion classification method

The invention provides an emotion classification method. The emotion classification method comprises the steps of obtaining a word embedding matrix corresponding to a context and a word embedding matrix corresponding to a target word; according to the word embedding matrix corresponding to the context, the word embedding matrix corresponding to the target word and the first semantic activation model, obtaining context representation with enhanced target word meaning and target word representation with enhanced context semantics; obtaining context representation after semantic selection according to the context representation of the target word semantic enhancement, the target word representation of the context semantic enhancement and the semantic selection model; according to the semantic integration model, extracting syntactic representation in a syntactic dependency tree corresponding to the target sentence; and obtaining an emotion classification result corresponding to the target word according to the context representation, the syntax representation and the second semantic activation model after semantic selection. Compared with the prior art, the semantic information related to the target word in the context is fully captured, and the relationship among the context, the target word and the syntax is comprehensively considered, so that the accuracy of sentiment classification is improved.
Owner:SOUTH CHINA NORMAL UNIVERSITY

False news detection system and method based on evidence perception hierarchical interaction attention network

The invention discloses a false news detection system and method based on an evidence perception hierarchical interaction attention network, which discusses more credible evidence semantics by capturing evidence fragments and fusing source credibility, and is used for discovering suspicious parts in false news so as to carry out interpretable false news detection. According to the method, suspicious parts in claim settlement can be captured, more credible evidence semantics can be explored for the parts, and the interpretability of false news detection is enhanced; the internal interaction module composed of self-attention and symmetric attention enhances deep interaction and matching between news semantics and related articles, so that key evidence fragments can be effectively obtained; the global reasoning module not only can obtain more credible evidences from key evidence fragments through gating affine transformation, but also can relieve the influence of extreme semantics on a detection result through global interactive reasoning.
Owner:XI AN JIAOTONG UNIV

Service discovery method combining attention mechanism LSTM and neural topic model

The invention discloses a service discovery method combining an attention mechanism LSTM and a neural topic model. The service discovery method comprises the following steps: 1, extracting natural language vocabularies from a label of a Web service description language and preprocessing the natural language vocabularies; 2, conducting semantic enhancement on the keywords extracted in the step 1, and obtaining described topic information through processing by the neural topic model; 3, embedding the keywords in the step 1 by utilizing a pre-trained word vector; 4, on the basis of the step 2 andthe step 3, carrying out described feature extraction by combining bidirectional LSTM of an attention mechanism; and 5, on the basis of the step 4, finding k services with the highest similarity froma registration service library by calculating the similarity of the feature vectors of the query request and the service description. The method has the advantages that semantic information existingin description of Web services is processed and enhanced through external information, and services meeting user function requirements are found in a large number of registration services on the basisof semantics inquired by a user.
Owner:WUHAN UNIV

Multi-video event blind area change process deduction method based on geographical semantic association constraints

ActiveCN112214642AEnhanced associative semanticsRealize quantifiable analysisSemantic analysisVideo data clustering/classificationProcess informationSemantic enhancement
The invention relates to a multi-video event blind area change process deduction method based on geographic semantic association constraint, which belongs to the technical field of geographic space data processing and comprises the following steps: a) extracting a longitudinal hierarchical structure and a transverse topological network of an enhanced semantic geographic position, and realizing geographic position semantic association expression under a unified positioning division benchmark of a monitoring scene interval; b) judging a geographic motion mode of a behavior-by-behavior process based on the relationship change characteristics of the track and the geographic position; c) establishing blind area behavior process characteristic parameters for mapping space-time distance migrationcost; and d) realizing blind area behavior process deduction through monitoring blind area geographic entity semantic path planning combined with a geographic position scene, a geographic motion modeand a space-time distance constraint. Geographic constraint deduction of blind area information in the discrete change process is completed, geographic semantic enhancement oriented to the continuouschange process of the monitoring area is achieved, and understanding of complete event change process information in different levels is supported.
Owner:SHENYANG INST OF APPL ECOLOGY CHINESE ACAD OF SCI

Modal independent retrieval method and system based on intermediate text semantic enhancement space

The invention discloses a modal independent retrieval method and system based on an intermediate text semantic enhancement space. The method comprises the following steps of obtaining the bottom features of pairs of text and image, constructing a training data set and a test data set; establishing the intermediate text semantic enhancement space based on linear discriminant analysis; calculating asimilarity matrix of image text pairs; respectively constructing a projection matrix model of an image retrieval text and a text retrieval image according to the intermediate text semantic enhancement space and the similarity matrix; using the training dataset to solve the projection matrix parameters of the retrieved text and the retrieved image respectively; and based on the projection matrix parameters of the text / text retrieval images, retrieving the text / text retrieval images by using the test data set. The method respectively learns mapping matrix for image retrieval text and retrievaltask of text retrieval image, the retrieval is more pertinent, and the retrieval accuracy is greatly improved.
Owner:SHANDONG NORMAL UNIV

Model training method, information display method and device

The invention discloses a model training method and an information display method and device. The method specifically includes obtaining a training sample from a pre-constructed sample set, and determining associated multimedia information corresponding to the target multimedia information; then, determining text features corresponding to the text information; performing semantic enhancement on the target multimedia information through the associated multimedia information to determine semantic features corresponding to the target multimedia information; on the basis of the features, determining the correlation degree between the text information and the target multimedia information through a to-be-trained matching model; and training the matching model according to the correlation degree and the label information corresponding to the training sample. Therefore, after the trained matching model is utilized, the multimedia information with the highest association degree with the text information can be selected as the multimedia information corresponding to the object described by the text information, so that a relatively high matching degree between the selected multimedia information and the text information is ensured.
Owner:BEIJING SANKUAI ONLINE TECH CO LTD

Intelligent voice dialogue method and device based on semantic enhancement, equipment and medium

The invention relates to the technical field of artificial intelligence, and discloses an intelligent voice dialogue method and device based on semantic enhancement, equipment and a medium, and the method comprises the steps: inputting to-be-recognized voice data into a preset voice recognition model, and carrying out the voice conversion of a text, and obtaining to-be-analyzed text data; inputting to-be-analyzed text data into the semantic enhancement text error correction model for semantic enhancement and error correction processing to obtain text data after error correction; inputting the text data after error correction into a preset intention recognition model for intention recognition to obtain an intention recognition result; according to the intention recognition result and the intention, performing matching with a verbal skill knowledge base to obtain target answer text data; and inputting the target answer text data into a preset speech synthesis model to carry out text-to-speech conversion so as to obtain target answer speech data. Semantic enhancement and error correction processing between text conversion and intention recognition are realized, and the accuracy of inputting the text of the preset intention recognition model is improved.
Owner:PINGAN PUHUI ENTERPRISE MANAGEMENT CO LTD

System for Physical-Virtual Environment Fusion

A semantic augmentation system includes a sensor with a computing system and a memory in communication with the computing system, the memory storing a plurality of endpoints. The computing system is configured to infer a first and a second semantic identity for an object, based on inputs from the sensor, project a coherent narrative and perform semantic augmentation towards a user. In further examples, the system infers a first narrative comprising two semantic identities and a second narrative wherein the system infers that a user observing view didn't infer the second semantic identity and further doesn't use the second semantic identity in the second narrative. It further, uses the corresponding narrative to remind the user to carry an item and / or credential in order to start an activity.
Owner:LUCOMM TECH

Visual saliency detection method based on semantic enhanced convolutional neural network

PendingCN110414513ASolve the problem that the deep features of the image cannot be extractedEnhance detailsCharacter and pattern recognitionNeural architecturesVisual saliencySelf adaptive
The invention discloses a visual saliency detection method based on a semantic enhanced convolutional neural network, and the method comprises the steps: improving a classic model VGG16, introducing aconvolution layer to replace a full connection layer, and better storing the detail information of an image; and adding a BN layer behind the convolution layer to accelerate the training speed of thenetwork, and adding a dropout layer behind the added convolution layer to solve the over-fitting problem of the network. And an SENet network unit is embedded after the final convolution layer to further improve the semantics of the network performance enhancement features. According to the method, the problem that deep features of the image cannot be extracted by a traditional method can be solved, detail information of the image can be enhanced, and loss of main feature information of the image and interference of noise in network propagation can be reduced by performing self-adaptive weighting on the extracted features. According to the invention, the visual saliency map with more accurate target area and less noise can be obtained.
Owner:UNIV OF ELECTRONICS SCI & TECH OF CHINA

Text abstract generation method based on feature extraction and semantic enhancement

The invention discloses a text abstract generation method based on feature extraction and semantic enhancement, and the method comprises the following steps: introducing a feature extractor, and obtaining a feature vector of an original text through the feature extractor; respectively connecting the feature vector with an output result of the encoder in a partial connection mode and a full connection mode, and filtering noise; acquiring long-distance dependence in the sentence by using a semantic enhancer, and further enhancing semantic association; using a convolutional neural network for carrying out feature extraction on a source sequence, enabling a feature extractor to directly act on a word vector of the source sequence, and meanwhile, keeping word vector layer parameters the same asword vector layer parameters of an encoder, so that the encoding process of the encoder and the feature extraction process of the feature extractor are ensured to act on the same semantic level. According to the method, the features of the sentence are extracted by using the feature extractor and then are further fused with the result of the encoder, so that the overall structure analysis of thesentence is facilitated, the noise in the text can be filtered, and the key information is found.
Owner:SHENYANG AEROSPACE UNIVERSITY

Zero-sample visual classification method for cross-modal semantic enhancement generative adversarial network

The invention discloses a zero-sample visual classification method for a cross-modal semantic enhancement generative adversarial network. The method comprises the following steps: firstly downloading a plurality of sample images and a tag and semantic features corresponding to each sample image, and extracting the visual features of each sample image through a residual neural network; and then constructing and training a cross-modal semantic enhanced generative adversarial network, and training a cross-modal feature classifier based on zero sample classification based on the generative adversarial network, thereby completing visual classification of the to-be-detected image.
Owner:UNIV OF ELECTRONICS SCI & TECH OF CHINA

Entity linking method and device based on semantic components

The invention provides an entity linking method and device based on semantic components, and relates to the technical field of computers. The method comprises the steps of obtaining a to-be-linked entity from a medical data set; determining a standard entity candidate set of the to-be-linked entities in a medical knowledge graph, wherein the standard entity candidate set comprises a plurality of standard entities most similar to the to-be-linked entities; determining a semantic similarity score of each standard entity in the candidate set and a to-be-linked entity through a pre-trained semantic enhancement model; and linking the entity to be linked to the standard entity with the highest semantic similarity score in the candidate set. Therefore, the semantic component information can be added into the entity linking process, so that the semantic enhancement model can learn more information, and the entity linking accuracy is higher.
Owner:BEIJING UNISOUND INFORMATION TECH +1
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Patsnap Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Patsnap Eureka Blog
Learn More
PatSnap group products