33results about How to "Rich semantic features" patented technology

The invention relates to a knowledge graph and discloses a knowledge graph representation learning method for integrating text semantic features based on an attention mechanism. The method solves theproblems that semantic features are insufficient due to the fact that a translation model does not utilize description texts of entities and relations, semantic features cannot be fused into entitiesand relations at the same time by a multi-source information embedding method, and the text extraction effect is poor. The method comprises the steps of firstly obtaining and processing description texts of entities and relationships to obtain text semantic features of the entities and the relationships, then constructing a projection matrix of the entities by utilizing the semantic features of the entities and the relationships, projecting entity vectors into a relationship space, modeling in the relationship space by utilizing a translation thought, and carrying out representation learning,so as to model a many-to-many complex relationship. The method is suitable for representation learning of the knowledge graph.

The invention discloses a Chinese text key information extraction method based on a pre-trained language model, which comprises the following specific steps of: (1) classifying key information to be extracted, easily concluding information categories forming rules, and extracting by using a regular matching method; and (2) extracting the named entities by using a sequence labeling model. (3) constructing the sequence labeling model by adopting a method of finely adjusting a pre-training language model, wherein firstly, a large-scale unlabeled text corpus is used for learning to obtain the pre-training language model, and word boundary features are introduced in a pre-training stage; (4) replacing the data content matched by using the rule with the corresponding rule template label so as tocomplete fusion of rule matching and the deep network; and (5) performing fine adjustment on the pre-trained language model according to the marked training data, and migrating the pre-trained language model to the sequence marking task of the named entity. According to the method, text context semantic features can be effectively extracted, and each information type can be effectively identifiedin a complex information type scene.

The invention relates to a video description method and system based on an information loss function, and the method comprises the steps: obtaining a training video, and obtaining the semantic information of each frame of a set training video; Inputting the semantic information of the training video into an LSTM-combined hierarchical attention mechanism model to obtain character description of thetraining video; According to the importance of each word in the character description to the expression video content, performing loss weighting on the words to obtain an information loss function, and taking the information loss function as an objective function to perform back-propagation gradient optimization on the hierarchical attention mechanism model to obtain a video description model; Obtaining a to-be-described video, respectively inputting the to-be-described video into the target detection network, the convolutional neural network and the action recognition network to obtain a setof target features, overall features and motion features of each frame of the to-be-described video as semantic information of the to-be-described video, and inputting the semantic information into the video description model to obtain character description of the to-be-described video.

The invention discloses a deformable convolution fusion enhanced streetscape image semantic segmentation method, which comprises a training stage and a test stage, and comprises the following steps of: constructing a streetscape image semantic segmentation deep neural network model to ensure that the network model obtains more small target feature information while a streetscape image large targetobject is segmented; therefore, the problems of small-scale target loss and discontinuous segmentation during streetscape image semantic segmentation are solved, the image segmentation effect is improved, the overall robustness of the model is better, and the streetscape image processing precision is higher.

The invention discloses an intelligent semantic matching method and device based on a depth feature dimension changing mechanism, and belongs to the technical field of artificial intelligence and natural language processing. The technical problem to be solved by the invention is how to capture more semantic context information and interaction information between sentences. Intelligent semantic matching of sentences is realized; the adopted technical scheme is as follows: the method comprises the following steps: constructing and training a sentence matching model consisting of an embedding layer, a depth feature variable-dimension coding layer, a convolution matching layer and a prediction layer; according to the method, deep feature variable-dimension coding representation of the sentences is realized, so that more semantic context information and interaction information between the sentences are obtained, and meanwhile, a convolution matching mechanism is realized, so that the purpose of intelligent semantic matching of the sentences is achieved. The device comprises a sentence matching knowledge base construction unit, a training data set generation unit, a sentence matching model construction unit and a sentence matching model training unit.

The invention relates to a text classification algorithm which combines statistical features and an Attention mechanism, and the Attention mechanism has been gradually applied to the field of naturallanguage processing. The prior method greatly increases the calculation amount when calculating the Attention weight value, and the invention proposes that the Attention weight value is calculated atthe structured event level. On the one hand, events contain richer semantics than words or phrases; On the other hand, the event-based Attention mechanism reduces computational complexity. At the sametime, statistical features are added to the calculation of Attention weights. Compared with the existing model, the semantic information contained in the event structure and the corresponding statistical features improve the quality of text vector representation and achieve better classification performance. The classification accuracy is evaluated, and the experimental results show that the model not only reduces the training time, but also achieves better results.

The invention belongs to the technical field of medical image processing, and relates to a pathological image analysis method based on deformation representation learning. The method comprises the following step that: a self-supervised deformation representation learning model is constructed, wherein the self-supervised deformation representation learning model is used for pathological image analysis and then used for classification and segmentation of a pathological image, wherein the learning model comprises a deformation module, a local heterogeneous feature sensing module and a global homogeneous feature sensing module, wherein the deformation module is used for performing elastic deformation operation on the image, the local heterogeneous feature sensing module is used for learning structural difference information caused by deformation of a local area in the image, and the module comprises a feature extractor network, a multi-scale feature network and a discriminator network. and the global homogeneous feature sensing module is used for realizing the learning process of the network on the global features of the pathological image. According to the method, the capability of extracting local structural features can be learned without marking data, and the global semantic information of the pathological image can be learned; compared with the best self-supervised learning method at present, the method of the invention is greatly improved in performance.

The invention discloses an intelligent question and answer oriented sentence pair semantic matching method based on a semantic feature map, and belongs to the technical field of artificial intelligence and natural language processing. The technical problem to be solved by the invention is how to capture more semantic context features, the relationship of coded information between different dimensions and the interaction information between sentences, and intelligent semantic matching of sentence pairs is realized. The adopted technical scheme is as follows: a sentence pair semantic matching model consisting of a multi-granularity embedding module, a deep semantic feature map construction network module, a feature conversion network module and a label prediction module is constructed and trained; and deep semantic feature graph representation of sentence information and two-dimensional convolutional encoding representation of semantic features are realized, and meanwhile, a final matching tensor of sentence pairs is generated through two-dimensional maximum pooling and attention mechanisms, and the matching degree of the sentence pairs is judged, so that the purpose of intelligent semantic matching of the sentence pairs is achieved. The device comprises a sentence pair semantic matching knowledge base construction unit, a training data set generation unit, a sentence pair semantic matching model construction unit and a sentence pair semantic matching model training unit.

The invention provides a question-answer text matching method based on artificial intelligence. The question-answer text matching method comprises the following steps: determining semantic similaritybetween a word vector sequence of a question text of a current user and a word vector sequence of each candidate question text in a candidate question text set through a deep learning model; determining the text similarity between the question text of the current user and each candidate question text in the candidate question text set through a preset formula; determining the comprehensive similarity between the question text of the current user and each candidate question text in the candidate question text set according to the semantic similarity and the text similarity; and according to thecomprehensive similarity between the question text of the current user and each candidate question text in the candidate question text set, determining a candidate answer text matched with the question text of the current user. In this way, the candidate answer text sent to the user is more accurate.

The invention discloses a software defect prediction method and terminal based on a bidirectional long short-term memory neural network, and the method comprises the steps: enabling an abstract syntax tree of a source code file and a source code to correspond to code change information between different versions through the bidirectional long short-term memory neural network; screening and extracting an abstract syntax tree point node sequence and a code change node sequence, connecting and constructing a combined sequence, inputting the combined sequence into a Word2Vec word embedding model, encoding the combined sequence into a word vector, fusing semantic features and traditional features by utilizing traditional measurement features provided by a PROMISE library and combining a gating fusion strategy to form combined features, and constructing a word vector; and inputting the combined features and the corresponding labels into a classifier to train a defect prediction model. According to the method, richer code semantic features are extracted from a source code abstract syntax tree and code change data, traditional features provided by a PROMISE storage library are combined, a classifier model is better helped to learn the semantic features, and a more accurate defect prediction result is obtained.

The invention provides a method and a device for detecting log sequence abnormity and a computer storage medium. The method comprises the following steps: collecting original log sequence data from a data source; sequentially extracting semantic information of each layer according to a hierarchical structure of a log sequence in the log sequence data, and generating a semantic vector of a fixed dimension of each layer; and calculating probability distribution of the log sequence by utilizing a SoftMax function according to the semantic vector of the log sequence, and selecting a result corresponding to the maximum conditional probability as an output category. According to the invention, richer semantic features can be extracted from words, logs and log sequences, so that the device can learn more accurate hierarchical semantic expressions, a better detection effect is achieved, and the capability of detecting three log sequence anomalies is further improved.

The invention is suitable for the technical field of computers, and provides a video classification method and device, a terminal and a storage medium. The method comprises the steps: obtaining a plurality of single-frame images corresponding to a to-be-processed video; inputting the plurality of single-frame images into a trained global feature extraction model for processing to obtain a first target image marked with global features and corresponding to each single-frame image; generating a first target video based on the at least one first target image; and inputting the first target video into a trained video classification model for processing to obtain a classification result corresponding to the first target video. According to the mode, the global feature corresponding to the single-frame image is extracted through the global feature extraction model, the target video is generated based on the image with the global feature, and when the target video is processed through the video classification model, the extracted semantic features are richer and more accurate; and when classification is carried out based on the semantic features, the classification result is more accurate, so that the accuracy of video classification is improved.

The invention provides an enterprise industry classification method based on a domain ontology. The method comprises the steps that (1) constructing a category feature word bank through the domain ontology, and forming a feature extension set used for extending feature words of a short text; (2) extracting category feature words from annotation information of National Economy Industry Classification Annotation (edition 2017) by using a TF-IDF feature extraction method for representing basic features of each industry category, and forming a feature corpus in a vector form; (3) extracting main business keywords of an enterprise, removing useless words, performing feature extension on the feature keywords by using the feature extension set of the domain ontology, and strengthening feature information of the feature keywords to obtain extended to-be-classified short texts; and (4) performing classification operation on the to-be-classified short texts subjected to feature extension by utilizing a BM25 classification model, and determining the industry category to which the to-be-classified texts belong according to the text similarity. The invention further comprises a system for implementing the enterprise industry classification method based on the domain ontology. The problems that time and labor are wasted and work is tedious in manual classification are solved.

The invention discloses an image zero-order classification model based on cross knowledge. The image zero-order classification model comprises: a biological classification tree module, which is used for constructing a biological classification tree according to all categories in the data set; the visual feature extraction module that is used for converting images in the data set into one-dimensional visual features; the semantic feature extraction module that is used for converting texts or attributes in the data set into one-dimensional semantic features; the cross knowledge learning module that is used for enriching semantic information of categories; the generative adversarial network module that comprises a generator and a discriminator, the generator generates pseudo visual features from the semantic features, and the discriminator is used for discriminating the authenticity and the category of the image. According to the invention, cross knowledge learning is adopted, more related semantic features can be trained, so that features from semantics to vision are embedded in the ZSL, and the semantic features in the cross-modal learning process are enriched; the model and the method are simple and efficient, and high-accuracy classification results are obtained on multiple authoritative data sets.

The embodiment of the invention provides a method and device for detecting an insulator image based on an insulator image detection model. The method comprises the steps of acquiring a target insulator image is acquired; performing feature extraction on the target insulator image through a first preset convolution layer to obtain a first feature map; performing spatial transformation on the firstfeature map to obtain a second feature map; performing feature extraction on the target insulator image to obtain a third feature map; processing the second feature map and the third feature map to obtain a fourth feature map; inputting the fourth feature map into a region candidate network, and outputting a fifth feature map; inputting the fourth feature map into a classifier, and outputting theinsulator type in the fourth feature map; according to the embodiment of the invention, the insulator of the insulator image can be accurately detected based on the improved feature extraction network, and the problem that the existing rich feature hierarchical structure FaterRCNN for accurate object positioning and semantic segmentation is low in insulator image detection accuracy is solved.

The invention provides a semantic similarity determination method. The method comprises the steps of obtaining a vector of a first target text and a vector of a second target text; respectively inputting the obtained vector of the first target text and the vector of the second target text into a recurrent neural network for encoding to obtain representation features of the first target text and representation features of the second target text; obtaining a similarity weight matrix of the first target text and a similarity weight matrix of the second target text according to the obtained representation features of the first target text and the representation features of the second target text; and determining the semantic similarity of the first target text and the second target text according to the obtained similarity weight matrix of the first target text and the similarity weight matrix of the second target text. Multi-dimensional semantic features can be obtained, and the accuracyof semantic analysis is improved.

The invention provides a gesture recognition method and device, electronic equipment and a computer readable storage medium, and the method comprises the steps: obtaining an original image which comprises a gesture; performing hand target detection on the original image to obtain a hand key feature data set; training a preset initial gesture recognition model by using the hand key feature data set as a training data set to obtain a basic gesture recognition model; pruning the basic gesture recognition model to obtain a standard gesture recognition model; and obtaining a to-be-recognized image, and recognizing a gesture type in the to-be-recognized image by using the standard hand recognition model. According to the gesture recognition method provided by the invention, the gesture recognition model is combined, the features of each gesture type are deeply mined, the trained gesture recognition model is pruned and optimized according to task requirements, and the accuracy of the model is improved on the premise of not additionally increasing computing resources.

The invention discloses a saliency target detection method and system based on local and global context fusion, and the method comprises the following steps: obtaining an RGB image to be subjected to saliency target detection, inputting the RGB image into a pre-trained saliency target detection model, and obtaining a saliency target detection result; wherein the saliency target detection model comprises a feature extraction network used for obtaining multi-scale features; the splicing fusion network is used for obtaining low-level detail features and high-level semantic features; the local context extraction network is used for outputting fine local context features; the global context extraction network is used for outputting fine global context features; and the fusion output network is used for outputting a saliency target detection result. According to the method, local and global context information can be effectively extracted, and the efficiency and accuracy of saliency detection in a complex scene can be improved.