232 results about "Label propagation" patented technology

The invention relates to the technical field of a social network and particularly relates to a multi-label propagation discovery method of overlapping communities in the social network. The multi-label propagation discovery method comprises the following steps: reading data of the social network, constructing a social network diagram which adopts social network users as nodes and user relationship as edges; according to the social network diagram, carrying out preliminary community division of the social network, and carry outing community discovery by adopting a label propagation method of comprehensively considering the node centrality and label-degree distribution constraint to obtain a non-overlapping community structure; marking the levels of the nodes according to the obtained non-overlapping community structure and the centrality value of the nodes in the communities; and according to the levels of the nodes, calculating label propagation gain among the nodes with different levels, and carrying out overlapping node mining by utilizing the multi-label propagation to obtain the overlapping community structure of the social network. The multi-label propagation discovery method has the advantages that the overlapping community structure in the social network can be effectively mined, the accuracy and the efficiency of community detection are favorably improved, and the method can be applicable to the fields of target group mining, precision marketing and the like.

The invention relates to a microblog-based neologism emotional tendency judgment method, belonging to the field of natural language processing. The microblog-based neologism emotional tendency judgment method disclosed by the invention comprises the following steps: dividing words of microblog corpuses through a Chinese word division tool, blocking the corpuses, the words in which are divided, by taking stop words in a word division result as a division point, pairwise combining adjacent word strings in each block, calculating the combined word string frequency, and taking the word strings, the frequencies of which are higher than a threshold value, as neologism candidate strings; filtering the neologism candidate strings according to a word formation rule of Chinese linguistics and an adjacent change number rule so as to obtain neologisms; calculating the similarity between co-occurrence words and hownet emotional words by utilizing an emotional dictionary of a hownet; calculating the relevancy between the neologisms and the co-occurrence words; constructing an image model; and obtaining the emotional polarity distribution of the neologisms by utilizing a label propagation algorithm, and obtaining the emotional tendency of the neologisms by constructing a linear classifier. By means of judgement of the emotional tendency of the neologisms, a blogger can express views better; and furthermore, the emotional tendency of the blogger can be accurately known by users.

The invention discloses a multilayer k-way graphical partition method based on vertex cut and community detection. The method comprises the steps that the distribution of a natural graph is considered according to the statistic analysis property, a corresponding vertex cutting algorithm is provided, vertexes causing longer task completion time are cut, label propagation is iteratively performed on the cut graph by a community detection algorithm based on the label propagation, the label of each vertex of the graph is determined, a community where the vertexes are located is obtained, partitioning is performed by a traditional multilayer k-way graph partition algorithm, and the efficiency is consolidated. For most of application in the large-scale iteration graph processing, distributed computational nodes meet the load balancing, extra communication traffic, due to the iteration dependency necessity, produced by each processing original node between adjacent iteration processing steps is greatly reduced, the task operating efficiency of a graph processing frame is greatly reduced, and the throughput capacity of tasks is increased.

Systems and methods for training semantic segmentation. Embodiments of the present invention include predicting semantic labeling of each pixel in each of at least one training image using a semantic segmentation model. Further included is predicting semantic boundaries at boundary pixels of objects in the at least one training image using a semantic boundary model concurrently with predicting the semantic labeling. Also included is propagating sparse labels to every pixel in the at least one training image using the predicted semantic boundaries. Additionally, the embodiments include optimizing a loss function according the predicted semantic labeling and the propagated sparse labels to concurrently train the semantic segmentation model and the semantic boundary model to accurately and efficiently generate a learned semantic segmentation model from sparsely annotated training images.

The invention discloses a fraud group identification method based on modularity and balanced label propagation. The method comprises: calculating the similarity degree between every two users of all users by using ID features in combination with the known fraud identifiers of the users, establishing a similarity matrix, and establishing an association diagram by using the similarity matrix; running a Louvain algorithm on the established diagram to obtain the community to which each node belongs and the hierarchical information of each node; using the community to which each node belongs, the hierarchical information of each node, and the fraud identifiers as the initial community information of each node, running a balanced label propagation process to obtain the community to which each node ultimately belongs, then dividing a network according to whether the nodes belong to the common community, and determining the fraud group according to the fraud identifiers obtained by the propagation. For the first time, the present invention applies the fraud group identification method based on modularity and balanced label propagation to the application anti-fraud and transaction anti-fraud fields. The method constructs the association diagram by using information such as transaction association, detects a fraud association by using a balanced label propagation algorithm in combinationwith the community modularity information, and prevents potential fraudulent transactions.

The invention relates to the technical field of social networks, in particular to a micro-blog hot word and hot topic mining system and method. The method includes the following steps that content data released in a micro-blog are preprocessed to acquire a candidate hot word sequence; according to the frequency of occurrence and suddenness of candidate hot words in a candidate hot word set at the current moment and in a given historical time window, the vitality of each candidate hot word is worked out, and a hot word set is formed by screening the candidate hot words; according to the hot word set formed by screening the candidate hot words, hot word correlation is worked out, and a hot word co-occurrence network is constructed; according to the hot word co-occurrence network, the hot word set is partitioned through the hot word clustering algorithm based on multi-label propagation to acquire a hot topic set. By means of the micro-blog hot word and hot topic mining system and method, efficient micro-blog hot word and hot topic mining is achieved, and mining precision and processing efficiency are improved.

Data are maintained in a distributed computing system that describe a graph. The graph represents relationships among items. The graph has a plurality of vertices that represent the items and a plurality of edges connecting the plurality of vertices. At least one vertex of the plurality of vertices includes a set of label values indicating the at least one vertex's strength of association with a label from a set of labels. The set of labels describe possible characteristics of an item represented by the at least one vertex. At least one edge of the plurality of edges includes a set of label weights for influencing label values that traverse the at least one edge. A label propagation algorithm is executed for a plurality of the vertices in the graph in parallel for a series of synchronized iterations to propagate labels through the graph.

The invention relates to an automatic image annotation method integrating depth features and semantic neighborhood. In view of the problem that manual selection of features takes time and energy in the traditional image annotation method, the problem that the traditional label propagation algorithm ignores semantic neighborhood, which results in visual similarity and semantic dissimilarity and affects the annotation result, and the like, the invention puts forward an automatic image annotation method integrating depth features and semantic neighbors. First, a unified and adaptive depth feature extraction framework based on a depth convolutional neural network (CNN) is built; then, a training set is grouped semantically, and a neighborhood image set of an image to be annotated is built; and finally, the contribution value of each label of the neighborhood images is calculated according to the visual distance, and the contribution values are sorted to get annotation keywords. The method is simple and flexible, and is of strong practicability.

Methods, systems, computer-readable media, and apparatuses for calendar matching of inferred contexts are described. In one potential embodiment, a mobile device may use context information to generate a calendar of inferred contexts. Label information from raw calendar data may be used to update an inferred context within a calendar of inferred contexts. Additionally, the label may be propagated to future contexts and entries in an inferred context calendar.

The invention discloses a type extracting system for a professional literature intellectual entity. The system comprises a user inquiring and feedback interface, an online crawler and management module, an intellectual entity identification module, an intellectual entity type extracting module, a type label propagation and index database establishment module, an intellectual entity type relational graph model establishment module, and a data visualization module. The system provided by the invention can be used for extracting the type of the entity according to key words of the entity searched by a user, and presenting a type relation and a hierarchical relation between intellectual entities and a timing sequence evolution mode in a visualization manner. Furthermore, the invention further provides a type extracting method for the professional literature intellectual entity. With the adoption of the method disclosed by the invention, type label extraction can be effectively carried out on the intellectual entities in professional fields; the problem of limitation and subjectivity in manual type predefinition can be solved; and the structured realization of a professional knowledge network is easily realized.

An integrated multi-classifier fusion classification method based on graph clustering label propagation comprises the following steps: using a training sample to train a basic classifier and clustering the training sample and a testing sample for multiple times to obtain multiple clustering partition states; carrying out label propagation based on the clustering partition states to obtain a clustering category label of the testing sample; processing all the clustering partition states and the basic classifiers according to the above-mentioned steps to obtain a clustering category information set of the testing sample; and making the clustering category information and classification information of the basic classifiers jointly constitute a decision matrix of an integrated classifier, setting parameters of a classification fusion target equation according to the clustering category information and the classification accuracy rate of the classification information of the basic classifiers so as to limit the range of the parameters in fusion, and using a BGCM method to carry out fusion classification on clustering category information of a to-be-classified sample and predicted label information of the basic classifiers according to the classification fusion target equation to obtain a final category label. The integrated multi-classifier fusion classification method is high in classification accuracy rate when difference exists among samples.

The invention relates to a remote sensing image road segmentation method based on convolutional neural network weak supervised learning. Sparse supervision information provided by road center line data is utilized, semantic features are propagated from a road center line to unmarked pixels through a context-aware label propagation algorithm, and a deep learning framework is combined to train convolutional neural network learning of a double-branch encoding-decoding structure to predict road surface data from a remote sensing image. The method has the advantages that the robustness is high, themethod can adapt to road surface segmentation of remote sensing images of different scales, continuous iteration and continuous optimization can be achieved, a road surface extraction result close tothe manual drawing level can be achieved only under weak label supervision, dependence on a large amount of training data of manual labeling is avoided, the labeling cost is greatly reduced, the method is an important step in automatic extraction of road research from remote sensing images, and has high application value in the aspects of resource exploration and planning, surveying and mapping,regional development and the like.

The invention provides a significance detection method based on sparse expression and label propagation. A new adjacent matrix is defined by use of the sparse expression, regions with common boundaries are called as adjacent regions and even data points in the same sub-space are defined as neighbors; then, a weight matrix is calculated through the similarity among regions in the image; parts of marginal regions are selected to be background labels; and label information of unlabeled regions is predicted through the weight matrix and the background labels obtained through the above method by applying a label propagation algorithm. The beneficial effects are that: by considering the relation between global information and local regions of the image, a new adjacent matrix is established; and by combining advantages of a sparse expression theory and the label propagation algorithm, the significance detection method has quite high accuracy and regression rate and is quite low in errors.

The invention discloses a human face recognition method and device based on tensor description. Firstly, similarity learning is conducted on image samples with labels and samples to be classified and without labels, and similar adjacent figures and normalized weights are configured to represent sample similarity; secondly, a category label matrix is initialized manually; thirdly, to effectively achieve direct induction of human face images outside samples, a regularization term capable of conducting direct induction on the images outside the samples and based on the tensor description is integrated into an existing label propagation model; finally, system modeling is finished through influences of parameter weighing similarity measuring, initial category labels and the regularization term based on the matrix pattern on the human face recognition; the maximum value of the probability of similarity in system output is taken to be used for conducting category identification of the human face images, and the most accurate system recognition result is obtained. By introducing the concept of tensor description, the topological structures among image pixels can be effectively maintained in the induction process of the human face images outside the samples, and the system expansibility is good.

A label propagation community structure mining method based on a node membership degree comprises the steps that a unique label is given to each node in a network, and each label is used for representing a community to which the corresponding node belongs; row vectors in a complex network adjacent matrixes are seen as sampling samples of the nodes, and a weight coefficient between two nodes is used as edge weight value; the variance of the node connecting edge weight coefficients is utilized as the node membership degree; in each label updating iteration, only the node labels with the membership degrees larger than the label updating threshold value are updated, and the nodes with the membership degrees smaller than the label updating threshold value are used as overlapping nodes; if the labels are changed, or the label propagation frequency is smaller than the label iteration threshold value, the iteration process is repeatedly executed, and if not, updating is stopped. The label propagation community structure mining method can well detect the overlapping community structure of the complex network under the situation that the time complexity increase is not large, and has the good robustness and accuracy.

The invention discloses a method utilizing Chinese online resources for supervising extraction of character relations remotely. According to the method, at first, an online encyclopedia website, formed through a semi-manual mode, on a web is utilized for automatically constructing a knowledge base so as to obtain accurate relation types comprehensive as much as possible, and examples of the character relations; then co-occurring names and context features are extracted from a text corpus, and the names and the relation examples in the knowledge base are matched to obtain name pair sets of the marked relations and name pair sets of unmarked relations; finally, a label propagation algorithm is introduced to achieve relation match of unmarked name pairs, so that extraction of the character relations is achieved. According to the method, the knowledge base of the character relations can be automatically constructed, and the richer and more accurate relation types are included; based on the knowledge base, the label propagation algorithm is introduced to supervise extraction of the character relations remotely, and therefore accuracy of results of the extracted relations can be ensured.

Method and system are described for generating a stratigraphic model of a subsurface volume. Measured geophysical data are converted into a vector volume (106) by assigning to each sample in each trace in the data volume a vector representing dip, azimuth or confidence. Then a labeled volume is generated from the vector volume by assigning a label to each sample in an initial trace (1006), then selecting a propagation pattern (1008) and propagating the labels to other traces (1010). Horizons can be extracted (110) from the labeled volume, and utilized to enhance the process of producing hydrocarbons (114).

The invention belongs to the technical field of network data mining and particularly relates to a detection method for an overlapping community based on multi-label propagation. The detection method comprises the following steps: A, constructing a social network diagram; B, analyzing a network rough core; C, initializing a label set; D, executing label propagation; E, decomposing a discontinuous community. According to the detection method disclosed by the invention, the link density between every two nodes is fully considered; the detection method has higher accuracy and effectiveness; in addition, manual input of data in the label propagation process is avoided.

The invention discloses a fraud call detection method. The method comprises the following steps that all call voices are converted into texts, thereby forming a text set; each text in the text set isconverted into a key word weight vector; a plurality of clusters are formed by performing text clustering on all key word weight vectors, and whether each cluster is a fraud cluster or not is determined according to a fraud keyword set; calls corresponding to all the keyword weight vectors in the fraud clusters are determined as fraud calls; a text social network is constructed by utilizing all the calls and keywords, nodes corresponding to the fraud calls are marked as the fraud calls in the text social network, and other nodes marked as the fraud calls are determined through label propagation; and the calls corresponding to all the nodes marked as the fraud calls are determined as the fraud calls. The method can be applied to various fraud types, and meanwhile, user sensitive data does not need to be acquired, so that the operability is higher.

The invention discloses a knowledge graph anti-fraud feature extraction method based on BFS and LPA, and the method comprises the steps: 1, carrying out the standardization of original data, converting the original data into labeled data under different dimensions, carrying out the cleaning and conversion, and forming data which conforms to the modeling of a knowledge graph; And step 2, constructing a knowledge graph model, including ontology construction, semantic annotation and information extraction. The method has the advantages that (1) a simple social relation is converted into a knowledge relation, different ontology knowledge is injected into a map, and a knowledge map representation method oriented to the consumer finance field is provided; (2) breadth-first search is introduced to find an entity touch black level, touch black information with different traversal lengths can be extracted after improvement, the feature level is enhanced, and the feature representation modes arediversified; And (3) for a fraudulent group problem in the anti-fraudulent field of consumer finance, entity subgroup information is mined by using an entity subgroup mining method based on label propagation, and a corresponding characteristic variable are extracted to show a relatively good distinguishing characteristic.

The invention provides a semantic disambiguation method for biomedical text polysemes. The method mainly comprises the steps of performing word vectorized representation on a biomedical text by utilizing Word2Vec; based on a bidirectional LSTM model, constructing vectorized representations of context sentences for a word vector language model; propagating existing labels for labeling medical datato most similar unlabeled data according to probabilities in combination with a label propagation method by utilizing a sentence vector space similarity relationship; and finally performing semantic disambiguation on the biomedical text in combination with all the labeled data. The biomedical data has the characteristics of strong speciality, numerous terms and the like, so that the operation of manually processing the medical data is time and labor-consuming and high in error rate; by using the method, the manual labeling cost can be greatly reduced; and compared with a conventional machine learning method, the semantic disambiguation accuracy can be effectively improved.

The invention belongs to the field of text classification, particularly relates to a short text labeling method, system and device for a large-scale classification system, and aims to solve the problem that the short text labeling system for the large-scale classification system is low in stability under the condition of limited data. The method comprises the steps that a first short text information set to be classified is acquired, and preprocessing is carried out based on a forward maximum matching segmented word and a word2vec word vector representation technology to obtain a second shorttext information set; based on a rule-based classification method and a supervised neural network classification method, perform binary classification on a second short text information set, then perform short text filtering, perform first-level and second-level classification labels of each short text based on the same classification method, and perform third-level and fourth-level classificationlabels of each short text based on a label propagation method of semi-supervised learning. According to the method, the stability of the short text label system oriented to the large-scale classification system is ensured under the condition of limited data.

The present invention provides a mobile communication user group construction method on the basis of fraction filtering and label propagation, belonging to the field of date business. The method comprises: calculating the connection closeness among user nodes; building a weighted complex network among the user nodes; constructing a seed group of a social relation group through adoption of a fraction filtering algorithm on the basis of the weighted complex network; initiating all the node labels in the network on the basis of the seed group, namely distributing initial labels for each node; and performing label propagation through adoption an improved SLPA algorithm, dividing the nodes to corresponding groups on the basis of the labels of the nodes when the labels of the major nodes are converged, and completing the construction of the social relation group, namely constructing a group of nodes which have the same labels. The mobile communication user group construction method on the basis of fraction filtering and label propagation is able to obtain a good user friend recommend effect and a good cooperation recommend effect, is helpful for users to form analysis and discover an abnormal colony, and is taken as the division base of the next generation communication.

The invention discloses a network community discovery method, which is used for community structure discovery in complex network analysis. The entire community discovery process of the present invention is divided into two stages. In the first stage, the node subgraph of each node in the network is used to divide the community into the community; in the second stage, based on the node subgraph The community partition results determine the order of node label propagation and update rules of each node community label. When propagating and updating the community labels of nodes in the network, the propagation order is no longer random, but the information entropy of nodes is used to measure the amount of information of network nodes, and the labels are propagated in the order of information entropy from small to large, so that Avoid the reduction of accuracy caused by randomness and uncertainty; at the same time, the update principle of node labels is no longer simply dependent on the frequency of labels, but comprehensively evaluates the uncertainty of node labels using optimal modularity and information entropy , thus avoiding the blindness of label updating.

The invention discloses a parallel label propagation-based heterogeneous network community discovery method. The method is characterized by comprising a label initialization step, a label cyclic refreshing step and a community building step. According to the label cyclic refreshing step, based on parallel label propagation, node labels are allowed to be propagated in multiple sub networks of the heterogeneous network in a relatively independent and parallel mode; and the node labels are refreshed through combining parallel propagation results of the multiple sub networks. Compared with a linear combination method LinearComb, the parallel label propagation-based combination method can use heterogeneous interactive information between nodes more effectively, and HLPA is more suitable for heterogeneous network community discovery.

The invention provides a parallel community discovery method based on label propagation under Spark, and relates to the field of data mining. In the invention, a complete subgraph is found in a network, and nodes in the complete subgraph are given the same label, so as to reduce the shortcomings of too many labels in an initial stage and improve the executing efficiency of an algorithm; then, according to the weight of the nodes, the propagation probability of the nodes in the network is calculated, the label propagation probability and the similarity among the nodes are considered in a label selection stage, and the accuracy of the label selection stage is improved; and the whole algorithm is executed in the framework of Spark, a good scalability is gained for the massive data, the executing efficiency and accuracy of the invention are both improved significantly, and the quality of community discovery is also greatly improved.

A method (100) and system (300) is described for processing video data comprising a plurality of images. The method (100) comprising obtaining (104, 106), for each of the plurality of images, a segmentation in a plurality of regions and a set of keypoints, and tracking (108) at least one region between a first image and a subsequent image resulting in a matched region in the subsequent image taking into account a matching between keypoints in the first image and the subsequent image. The latter results in accurate tracking of regions. Furthermore the method may optionally also perform label propagation taking into account keypoint tracking.

The invention discloses a book body matching method based on machine learning, and mainly aims to solve the problem of difficulty in realizing body matching in the field of book processing. The method comprises the following steps: generating all instance pairs and concept pairs to be matched for two given book bodies, and mining equivalence relations between instances from all the instance pairs to be matched by using a heuristic instance matching rule and a decision model based on supervised learning in order to obtain an instance matching result; mining hyponymy relations and equivalence relations between concepts for all the concepts to be matched by using a label propagation algorithm based on semi-supervised learning in order to obtain a concept matching result; finally, combining the instance matching result and the concept matching result for serving as a book body matching result.

A method (100) and system (300) is described for processing video data comprising a plurality of images. The method and apparatus is for obtaining for labelling of a plurality of objects or regions in an image of a sequence of images followed by label propagation to other images in the sequence based on an inference step and a model.

The invention relates to a label propagation community discovery algorithm based on node importance, and its main technical features are: initializing the unique label of each node; calculating the importance of each node, and sorting the nodes according to the node importance from high to low, Generate an ordered sequence; set the number of iterations t=1; for any node in the ordered sequence, update the label of the node to the label with the greatest influence in the label set of adjacent nodes according to the label selection method and label update rule; if the number of iterations t==max Iter or the label of each node is the most influential label, then the nodes with the same label are classified into the same community, and the process ends; otherwise, the number of iterations t is increased by 1, and the update is continued. The present invention has reasonable design, can significantly improve the quality of community discovery under the condition of similar complexity, shorten the iteration cycle, has high accuracy and stability, and can be widely used in community discovery, social network and other fields.