44 results about "Relational reasoning" patented technology

The invention discloses a point cloud semantic segmentation method based on point global context relation reasoning. The method comprises the steps of obtaining a training set T and a test set V; constructing a point cloud data semantic segmentation network of deep learning and global context reasoning; using a multi-classification cross entropy loss function as a loss function of the point cloudsemantic segmentation network; performing P rounds of supervised training on the point cloud data semantic segmentation network by using the training set; and inputting the test set into the trained network model for semantic segmentation to obtain a segmentation result of each point. The method has the beneficial effect that the problem of insufficient global information extraction of 3D point cloud semantic segmentation is solved by utilizing a method based on deep learning and global context reasoning. On the basis of deep learning, an added global context inference module models the relationship between feature channels by using a channel attention mechanism, global information of the relationship between the channels is further transmitted and aggregated through graph convolution, andthe global information can be obtained so as to refine the result of point cloud semantic segmentation.

The invention discloses an information system management method based on a knowledge graph. According to the method, multi-source heterogeneous data such as technical documents, test analysis reports,operation log data and user feedback generated in design and actual operation of an enterprise informatization system are utilized, entities and relations in the data are extracted by comprehensivelyutilizing iteration rules and a machine learning algorithm, and a knowledge graph of the enterprise information system is constructed. On the basis of the extracted triple data, logic rules and machine learning are combined to realize relationship reasoning of the knowledge graph, a hidden relationship of an enterprise internal information system is mined, existing knowledge graph data is supplemented and perfected, and finally, a webpage is designed by utilizing a B / S mode to visually display the knowledge graph of the enterprise internal information system. The knowledge graph visually andcomprehensively displays the composition structure of each information system of the enterprise, helps enterprise decision makers and technical management personnel to macroscopically master the interaction relationship of the internal information systems of the enterprise, and promotes healthier and more stable development of internal informatization construction of the enterprise.

The invention discloses a knowledge map relation inference algorithm based on a stacked neural network, and belongs to the technical field of artificial intelligence representation learning. The invention constructs a stacked neural network model comprising two components: a standard LSTM loop neural network and a multi-layer perceptron network. The model regards the triplets in the knowledge mapas short sentences and uses the learning ability of LSTM loop neural network to model the logical and semantic characteristics of the knowledge map in order to learn the grammatical and semantic information in the knowledge map. Through the bottom-level feature learning process, the upper-level fully connected network can provide enough discrimination to distinguish objects in different contexts.The algorithm of the invention models the three tuples in the knowledge map from the perspective of semantics, fundamentally solves the semantic diversity of entities and relationships, can greatly reduce the computational overhead of the relational reasoning of the large-scale knowledge map, and simultaneously can ensure the effectiveness of the relational reasoning.

The invention provides a knowledge graph relation reasoning method based on a graph neural network, and the method comprises the steps: mainly dividing a relation reasoning model in knowledge graph reasoning into a scoring function and a prediction result, wherein a candidate result set composed of the target node and the relationship can be obtained according to different modes according to the result of the relationship reasoning. In the process of reasoning the correct relationship in the candidate result set, the method mainly combines the structure information in the knowledge graph with the semantic and attention mechanism of the knowledge graph to realize the reasoning prediction of the unknown relationship in the knowledge graph. In the process of obtaining semantic information of the knowledge graph, structural information around target head and tail entities is extracted by using an isomorphic information theory, attention information collection is performed on the knowledge graph around a target relationship, and a candidate result with the highest score is output as a final result through effective fusion of an attention mechanism.

The invention provides a multi-source heterogeneous knowledge graph collaborative reasoning method and device based on entity alignment, and the method comprises the steps: determining an equivalent relation path of a query relation based on a to-be-reasoned entity pair and the query relation of the to-be-reasoned entity pair; and inputting the feature vectors corresponding to the equivalent relation paths into a relation reasoning model to obtain a reasoning result corresponding to the query relation. The positive sample triples used for training the relational reasoning model are obtained from the multiple knowledge maps, meanwhile, the equivalence relation paths comprise the equivalence relation paths crossing the knowledge maps, communication between different knowledge maps is achieved, and therefore the relational reasoning model can learn semantic information in the multiple knowledge maps, and the problem of relatively low reasoning precision caused by knowledge reasoning for a single knowledge graph in a traditional method is avoided.

The invention discloses a relation reasoning method based on a deep neural network. The method comprises the steps of after obtaining sample sentences, constructing a syntactic dependency tree consisting of a plurality of words according to a preset fusion rule, then respectively extracting a main feature of the syntactic dependency tree on a shortest dependency path and an auxiliary feature on anon-shortest dependency path, finally carrying out feature fusion on the main feature and the auxiliary feature according to the preset fusion rule, and obtaining a relation reasoning result accordingto the fusion result. Visibly, according to the method, the characteristics of the syntactic dependency tree on the shortest dependency path and the non-shortest dependency path are extracted respectively and fused, and due to the fact that the auxiliary characteristics have a certain auxiliary effect on the reasoning result, the accuracy of relation reasoning is remarkably improved by effectively utilizing the main characteristics and the auxiliary characteristics of the syntactic dependency tree. In addition, the invention further provides a relation reasoning device and equipment based onthe deep neural network and a computer readable storage medium, and the effects of the relation reasoning device and equipment correspond to the effects of the method.

The invention discloses a complex network link prediction method and system based on logical reasoning and graph convolution. The method comprises the following steps: constructing a knowledge graph corresponding to a complex network, and obtaining a training set; performing relation reasoning on each entity pair in the training set through a first-order logical reasoning network with default, and obtaining a relation confidence coefficient matrix through mapping; based on the relation confidence coefficient matrix, performing iterative training on a graph convolutional neural network based on iterative attention through a centralized training decentralized execution mechanism and a local relation attention mechanism to obtain first probability distribution; calculating second probability distribution according to a relation weight matrix and a relation confidence coefficient matrix output by network iteration; obtaining a Wasserstein distance between the first probability distribution and the second probability distribution according to a joint evaluation function; iteratively updating the two networks according to a Wasserstein distance to obtain a link prediction model; and complementing the knowledge graph according to the link prediction model. The link prediction efficiency is high.

The embodiment of the invention provides a method and device for constructing a network security knowledge graph, and relates to the technical field of network security. The construction method of thenetwork security knowledge graph comprises the steps: firstly obtaining original relation data for constructing the network security knowledge graph, and constructing an original relation graph according to the original relation data; then performing relationship reasoning processing on an original relationship graph through a pre-constructed relationship reasoning model to obtain a missing relationship set; and finally, performing optimization processing on the original relationship graph according to the missing relationship set to obtain a network security knowledge graph. Visibly, in themethod, the missing relation can be automatically reasoned, the network security knowledge graph can be generated, manual participation is not needed, the accuracy is high, the speed is high, and theconstruction efficiency of the network security knowledge graph can be improved.

The invention provides a self-supervised learning model training method and device based on relational reasoning. The method comprises obtaining different local observation images corresponding to theimages through different geometric transformation operations; extracting local features corresponding to the corresponding images; fusing the local features to obtain global features of the corresponding images; predicting a corresponding prediction geometric transformation operation between the local feature and the global feature; according to the difference between the prediction geometric transformation operation and the actual geometric transformation operation; constructing a loss function of the learning model; determining target parameters of the learning model through iteration of the loss function; using the prediction geometric transformation operation as a supervision signal to train a learning model. The relationship of the preset auxiliary task is established between the global feature and the local feature, so that the feature obtained by model learning can focus on capture of semantic information of the visual object, the influence of the preset auxiliary task on feature learning is reduced, and migration to the target task is easy.

The invention provides a machine reading inference method based on a graph neural network. Overall process is as follows: a proposition judgment module, an entity identification module and an entity chain finger module are obtained through secondary training of a neural network; an information extraction module and a polarity discrimination module are combined respectively; a fact logic relation graph in a reading material and entity and polarity information in a to-be-inferred proposition are obtained, and then the fact logic relation graph, together with an environment knowledge graph, is input into a graph neural network subjected to secondary training together to obtain a final entity logic relation graph; and finally an inference conclusion and an inference route graph are obtained byusing a Bayesian network. According to the method, the graph neural network is applied to machine reading inference for the first time; on the basis of relation inference, the machine logic inferencecapacity is further given, and the automatic case inference process is achieved; and the method has important use value in the fields of criminal investigation, machine questioning and answering andthe like.

The invention relates to a relation prediction method combining logic rules and fragmented knowledge. The method comprises the following steps: firstly, performing unified modeling on a fact triple and a logic rule, and embedding hidden semantic information into a knowledge representation-based relationship reasoning model; secondly, in combination with fragmented knowledge, performing continuousiterative updating, so that a knowledge base becomes more complete. According to the method, unified modeling is carried out on the fact triad and the logic rule, hidden semantic information is embedded into the relation reasoning model based on knowledge representation, and more accurate prediction is achieved.

The invention relates to the technical field of visual questions and answers, discloses a visual questioning and answering method based on a mathematical chart data set, and provides the following scheme for existing five types of mathematical chart visual questions and answers: S1, performing target detection based on a Faster-RCNN model; wherein the model is used for identifying object instancesbelonging to certain classes and determining the position of the object in the image through a bounding box; s2, performing object feature coding based on a target detection result; s3, preprocessingthe question-answer pair; s4, performing the text question feature extraction; and S5, performing visual intelligent questioning and answering based on a relation reasoning module, in S1, target detection based on a Faster-RCNN model comprises two stages of detection objects. The invention provides a solution combining a target detection module and an inference module, which is used for solving the visual question-answering problem of five types of mathematical charts, and improves the accuracy.

The invention discloses a knowledge graph reasoning completion method and system based on reinforcement learning, and the method comprises the steps: loading data, analyzing the data, and obtaining the vector representation of entities and relationships of a knowledge graph; embedding the entities and the relationships on the basis of a TransE model, and pre-training a pre-constructed strategy network to obtain a pre-trained strategy network; retraining the pre-trained strategy network through a plurality of reward functions to obtain a trained strategy network; and inputting the knowledge graph to be tested into the trained strategy network to complete the completion of the knowledge graph. The system comprises an analysis module, a pre-training module, a re-training module and a completion module. Through the design of the reward function, the complemented relation reasoning path is more reliable and has diversity. The knowledge graph reasoning completion method and system based on reinforcement learning can be widely applied to the field of knowledge graphs.

The invention provides an interpretable reasoning question-answering method and device. The method comprises the steps of constructing a knowledge graph of related fields of a to-be-answered question;converting the knowledge graph into a graph structure, and predicting a semantic relationship between entities in the graph structure; based on the semantic relationship, obtaining a logic rule between entities; complementing the knowledge graph based on the logic rule; and according to the to-be-answered question, obtaining an answer to the question based on the completed knowledge graph. Relational reasoning is carried out by applying rules, and the goal of knowledge graph completion is achieved, so that the interpretability of the reasoning process is improved, a question-answering systemanswers more accurately, and the user experience is improved.

The invention discloses a knowledge graph construction method and device based on UCL semantic indexing. The method comprises the following steps: firstly, extracting entities and relationships from an open link database, and constructing a basic entity library; then crawling news information in each portal website by utilizing a crawler, cleaning and analyzing the news information to form a UCL entity, completing entity disambiguation, and carrying out semantic fusion on the UCL entity and the basic entity library to construct a knowledge graph; and finally, completing the automatic completion of the knowledge graph by utilizing an improved relation reasoning algorithm, so that the automation level of the knowledge graph is improved. According to the method, standardized and fine-grainedsemantic indexing can be carried out on the internet information by utilizing the UCL national standard, so that semantic information expressed by the knowledge graph is richer, and knowledge graph support based on UCL semantic indexing is provided for content big data application based on semantic deep processing.

The invention discloses a qualitative spatial relation reasoning method based on rules and constraint satisfaction.Reasoning is performed based on the rules, and possible spatial relations between objects are reasoned; the thought of a constraint-satisfaction problem is used for judging the consistency between the possible spatial relations, and a result which will cause a conflict is removed through a route consistency method so that the consistency between the possible spatial relations can be guaranteed; possible spatial relations generated after removal are reserved, and therefore spatial relation reasoning can be achieved.The qualitative spatial relation reasoning method based on the rules and constraint satisfaction can verify the consistency between results, and possible defects of the reasoning method are avoided.More importantly, the method can guarantee consistency between the reasoning results, and compared with existing methods, reasoning efficiency is improved to a certain extent.

The invention discloses a migratable medical inquiry dialogue system and method for a low-resource scene, and the system comprises: a simulation patient construction module which is used for buildinga simulation patient which is provided with a plurality of dialogue samples, randomly selects one dialogue sample for each training, and gives an illness state description report for the simulation patient; an intelligent medical inquiry system construction module which is used for establishing an intelligent medical inquiry system, extracting context information of different hierarchies from dialogue history through the layered context encoder module according to disease description, encoding to obtain feature vectors of single-round hierarchies and multi-round hierarchies of dialogues, and establishing an intelligent medical inquiry system by evolving an external medical knowledge graph, relationship reasoning between disease symptoms is performed according to the feature vectors to obtain graph node feature vectors to generate replies of doctors to patients by utilizing a replication network under the guidance of graph node information; and a training module which is used for carrying out end-to-end training on the system and training the system on dialogue data of existing diseases by utilizing a graph evolution meta-learning algorithm.

The present disclosure discloses a visual relationship detection method based on adaptive clustering learning, including: detecting visual objects from an input image and recognizing the visual objects to obtain context representation; embedding the context representation of pair-wise visual objects into a low-dimensional joint subspace to obtain a visual relationship sharing representation; embedding the context representation into a plurality of low-dimensional clustering subspaces, respectively, to obtain a plurality of preliminary visual relationship enhancing representation; and then performing regularization by clustering-driven attention mechanism; fusing the visual relationship sharing representations and regularized visual relationship enhancing representations with a prior distribution over the category label of visual relationship predicate, to predict visual relationship predicates by synthetic relational reasoning. The method is capable of fine-grained recognizing visual relationships of different subclasses by mining latent relationships in-between, which improves the accuracy of visual relationship detection.

The invention provides a relation reasoning method and system based on a dependency graph. The relation reasoning method comprises the following steps: performing word division and word feature construction on a given sentence pair by using word meaning features; obtaining a dependency relationship tree between words extracted from the text after word division through a dependency extractor; taking the dependency relationship as a basis for word feature updating, and learning and updating word features in a given sentence pair in combination with a deep learning network; taking the plurality of updated word features obtained by the given sentence pair as local features, and performing feature fusion to obtain global features; performing interaction between two sentences by taking the global features as sentence meaning features, inputting the sentence meaning features into an output layer to obtain output, comparing the output with a real label, and calculating a loss function of a learning model; and correcting the learning model according to a loss function calculation result of the learning model, and determining a target parameter corresponding to the learning model. And the expression of the syntactic dependency tree on natural language reasoning is effectively improved.

The invention discloses a hyperspectral classification method of a lightweight hybrid convolution model based on global reasoning, belongs to the technical field of information processing, and is mainly used for hyperspectral image classification of small sample data. The lightweight hybrid convolution model based on global reasoning comprises a layer of two-dimensional convolution, a layer of three-dimensional convolution and a global reasoning module; a global reasoning module is added, global feature information and deep feature information of the hyperspectral image are effectively extracted by reasoning context relationships among different regions, feature extraction of deep three-dimensional convolution is replaced, and the complexity and calculation cost of the model are greatly reduced. A test result in a public data set shows that the classification performance of the method is superior to that of the current best classification method, the space-spectrum joint features of the hyperspectral image can be effectively extracted only by a small number of training samples, and the method is high in practicability. The problem that channel relation information is lost due to the fact that only two-dimensional convolution is used, and the problem that model complexity and calculation cost are greatly increased due to the fact that deep three-dimensional convolution is adopted are solved.

The invention provides an intelligent question-answering method and system based on a knowledge graph and a knowledge graph updating system. The method comprises: receiving a user question; convertingthe user question into a regularization question; and performing question matching on the regularized questions and a template library generated by a knowledge graph based on a heterogeneous information model, and determining a matching result with the highest similarity as an answer. Characteristics of a heterogeneous information network are fully utilized for modeling, richer knowledge and relationship reasoning are provided by combining a ranking collaborative clustering algorithm and a meta-path similarity algorithm, a traditional knowledge graph-based question-answering system is improved, and the accuracy and the intelligent degree of the question-answering system are improved.

The invention provides a knowledge graph complex relationship inference method based on multidirectional semantics. The method comprises the steps of mapping entities in a training sample data set of a knowledge graph into two groups of low-dimensional space vectors for representation; mapping a relationship in a training sample data set of the knowledge graph into two groups of low-dimensional space vectors and one-dimensional parameter representations; entities in the training sample data set of the knowledge graph are randomly selected to replace entities of the training sample positive triad to generate training negative sample data; defining an objective function in the training process according to the training sample positive triad and the generated training negative sample; respectively substituting an entity mapping result and a relationship mapping result in the training sample data set into the target function, and optimizing to obtain vector representation corresponding to each entity or relationship in the knowledge graph; and calculating a distance value between the entity and the relationship in the knowledge graph triad by utilizing the vector representation obtained by optimization, and performing relationship reasoning according to the distance value. According to the method, the reasoning effect on the complex relationship is improved.

The invention discloses a text detection method, device and system based on depth relation reasoning and a storage medium, and the method comprises: obtaining a to-be-detected text image, and carrying out the geometric attribute estimation of a rectangular component in the to-be-detected text image through a pre-constructed and trained text component network, wherein the text component prediction network adopts a convolutional neural network in cross-layer connection; generating a plurality of local graphs according to the geometric attributes of the rectangular components; and performing deep reasoning on the local graph through a pre-constructed and trained deep relation reasoning network, and forming a text detection result according to a reasoning result link. According to the embodiment of the invention, the local graph is generated after the geometric attributes of the rectangular components in the to-be-detected text image are estimated, and the deep relation reasoning is further executed for the local graph to establish the link between the rectangular components so as to obtain the text detection result. The stable relation between the component areas is mined by utilizing depth relation reasoning, so that the detection performance of the text in any shape can be greatly improved.

The invention discloses a relation reasoning method, a relation reasoning model obtaining method and device and a storage medium. The relation reasoning method can comprise the steps: obtaining presetinformation of all personnel serving as reasoning objects; constructing a directed graph according to the predetermined information; for the two persons serving as inference objects, generating relation characteristics of the two persons according to the directed graph; and inputting the relationship characteristics into a pre-trained relationship reasoning model to obtain a reasoned social relationship between the two persons. By applying the scheme provided by the invention, the accuracy of the reasoning result and the like can be improved.

The invention relates to a multi-hop knowledge graph question answering method and system based on knowledge graph embedding, and belongs to the technical field of natural language processing application. According to the method, firstly, a knowledge graph is embedded, and whether a triple exists or not is judged by calculating the score of a vector combination of the triple (a head entity, a relation and a tail entity); for a multi-hop problem, attention calculation is carried out by utilizing an attention mechanism and all potential relationships with a problem node, and a relationship code with a relatively high weight is fused into a code of the problem, so that the problem code is not generated by voucher, but is obtained by reasoning and calculating depending on a known relationship of a map. Meanwhile, on the basis that the known relation can be deduced, question nodes are inferred reversely through answers, the accuracy of the answers is verified, and the accuracy of the answers is further improved.

The invention relates to the field of cross-modal retrieval, and discloses a video and text cross-modal retrieval method based on a relational reasoning network, which comprises the following steps: extracting video data features and text data features; obtaining video global features and text global features by using a recurrent neural network; constructing video local relation features and text local relation features by using a multi-scale relation reasoning network; respectively fusing the global features and the local relation features of the single-mode data to obtain video fusion features and text fusion features; mapping the video fusion features and the text fusion features to a public space, and aligning video fusion feature distribution and text fusion feature distribution in the public space; and training the whole network. According to the method, global features and local relation features are concerned at the same time, key information in single-mode data can be more effectively focused, and then cross-mode retrieval is achieved.

The present disclosure discloses a visual relationship detection method based on adaptive clustering learning, including: detecting visual objects from an input image and recognizing the visual objects to obtain context representation; embedding the context representation of pair-wise visual objects into a low-dimensional joint subspace to obtain a visual relationship sharing representation; embedding the context representation into a plurality of low-dimensional clustering subspaces, respectively, to obtain a plurality of preliminary visual relationship enhancing representation; and then performing regularization by clustering-driven attention mechanism; fusing the visual relationship sharing representations and regularized visual relationship enhancing representations with a prior distribution over the category label of visual relationship predicate, to predict visual relationship predicates by synthetic relational reasoning. The method is capable of fine-grained recognizing visual relationships of different subclasses by mining latent relationships in-between, which improves the accuracy of visual relationship detection.

The invention discloses a relation mining method and system based on graph structure data, and the method comprises the steps: S1, obtaining and analyzing an image, and obtaining the graph structure data; S2, processing the graph structure data based on a task layer, and performing relationship reasoning on a graph structure to obtain sub-tasks; S3, completing interaction with the environment according to the sub-tasks to obtain corresponding rewards; S4, feeding back the corresponding rewards to the task layer; and S5, circulating the step S2-S4 until the sub-task with the maximum reward is completed. The system comprises an object and relationship detection module, a task layer module, an action layer module, a feedback module and a circulation module. According to the method, modeling is directly carried out, the relation between objects is utilized, and interpretation performance can be provided when the same performance is achieved. The relation mining method and system based on graph structure data can be widely applied to the field of reinforcement learning.

The invention discloses a complex network link prediction method and system based on logical reasoning and graph convolution. The method includes: constructing a knowledge graph corresponding to a complex network, and obtaining a training set; performing relational reasoning on each entity pair in the training set through a default first-order logical inference network, and obtaining a relational confidence matrix through mapping; based on relational confidence degree matrix, iteratively trains the graph convolutional neural network based on iterative attention through the centralized training decentralized execution mechanism and the local relationship attention mechanism, and obtains the first probability distribution; calculates according to the relationship weight matrix and relationship confidence matrix output by the network iteration The second probability distribution; obtain the Wasserstein distance between the first probability distribution and the second probability distribution according to the joint evaluation function; iteratively update the two networks according to the Wasserstein distance to obtain a link prediction model; complete the knowledge graph according to the link prediction model. The link prediction efficiency of the present invention is high.