62 results about "Network mining" patented technology

Described is a system for supporting human intelligence analysis. The system detects changes in social relations among users within a dynamic information network and enables understanding of a current social situation in the dynamic information network through multiple integrated modules. An active network mining module identifies incomplete data that is related to at least one change in the social relations and resolves conflicting and missing data in the dynamic information network. A relevant network discovery module constructs a relevant network from hidden relations within the dynamic information network. An information-aware social network module constructs an information-aware social network using the relevant network, then classifies and prioritizes items of interest to provide an assessment of a current social situation to a user.

The invention discloses a remote sensing image building extraction method and system based on a U-Net network, and electronic equipment. A multi-scale module is added to a decoding layer of a U-Net network, and the hole convolution network is introduced, the receptive field can be expanded under the condition that the resolution is not lost through hole convolution, so that the semantic information mining capacity of the network can be improved while detail information is reserved, and meanwhile, the multi-scale feature obtaining capacity of the network is enhanced through the multi-scale module; according to the invention, the convolution mode of the convolution layer is set as filling; that is, after convolution, the size of the feature map is completely unchanged; the original feature map is actually shrunk by 2; in this way, each time the feature map passes through a convolution layer , the size of the feature map is reduced by two times; by the adoption of the convolution model, the size of the feature map output through the four coding layers and the last coding layer is shrunk to be one sixteenth of the size of the input picture after the feature map passes through 4 encoding layers, the image resolution is recovered through deconvolution operation, the size of the feature map begins to be enlarged at the moment, and the training time is effectively shortened.

The invention discloses a driver attention area prediction method and system based on target dynamic information. The method comprises the steps that spatial features of video frame images and dynamicfeature maps of adjacent video frame images are extracted; important target screening is carried out on targets in the extracted video frame images, cross-scale fusion is carried out on target feature maps of different scales, and cross-scale target features are obtained; attention fusion is performed on the spatial features and the cross-scale target features, and a driver attention prediction network model is trained with the dynamic feature map; and the trained driver attention prediction network model is adopted to predict the driver attention area of the to-be-tested video frame image. Through an important target screening network, an important target possibly existing at the current moment is mined, and the important target is fused with the image spatial features to enrich the spatial expression ability of the model; the inter-frame dynamic information is extracted through the extraction of the dynamic feature map, so that the method is more sensitive to the motion informationof an important target, and the prediction precision of the attention of a driver is improved.

Methods, systems, devices and / or apparatuses are provided for computationally deriving molecular association connectivity maps for the study of inter-class molecular associations in toxicogenomics and drug discovery applications. The inter-class molecular associations can be between at least one bio-molecular entity and at least one therapeutic agent. The methods, systems, devices and / or apparatuses apply integrated molecular interaction network mining and text mining techniques.

According to the cross-modal retrieval method and system based on semantic condition association learning, the multi-label information serves as a new observation mode, and the multi-label semantic relation is effectively integrated into a cross-modal implicit representation learning framework based on a deep neural network. On one hand, the feature learning process of each mode is guided throughlabel semantic information, depth feature representation which keeps a semantic relation and has discrimination ability is obtained, and the cross-modal retrieval performance is improved; on the otherhand, the high-level semantics in the multi-label data are mined by using the deep network, and the typical correlation of different modal features with respect to the high-level semantics is maximized by using a conditional association learning method, so the shared semantic information can be eliminated from each modal data, and a direct association relationship between different modals is established. The method is advantaged in that the influence of the noise label on the cross-modal implicit representation is effectively reduced.

The invention discloses a mining Trojan horse detection system based on flow analysis, which relates to the field of computer network security and comprises a connectionmining pool mining Trojan horsebehavior detection subsystem and a p2p mining network mining Trojan horse behavior detection subsystem. According to the invention, the static pcap data packet or the real-time flow is used as the input, two modes of detecting and connecting mine pool mining and p2p mining can be selected, the mining flow is analyzed through the field feature extraction or communication flow feature extraction and identification of the system, and the alarm information is output to the user. Aiming at the conditions of plaintext communication and ciphertext communication, the system has the capability of quickly processing mass data, and can meet the requirements of personal hosts and enterprise-level users at the same time.

The invention provides a multi-modal remote sensing image data detection method and system, electronic equipment and a storage medium, and the method comprises the steps: inputting multi-modal remote sensing image data of different time sequences into a multi-modal feature mining network, and outputting the fusion vector features of the multi-modal remote sensing image data of each time sequence, and inputting the plurality of fusion vector features into a change detection network, and identifying whether the multi-modal remote sensing image data with different time sequences have differences or not by the change detection network. The multi-modal feature mining network constructed by the invention is used for mining and fusing the features of the multi-modal remote sensing image data, so that the diversity, multi-time sequence and multi-feature requirements of a training data set are met, and the accuracy of network feature mining can be improved; and the abnormal states of the multi-modal remote sensing images with different time sequences are detected by constructing a multi-modal feature mining network and a change detection network, so that support is provided for researching the development trend of a detected area.

The invention discloses an overhaul period optimization method for an intelligent substation protection system, and the method comprises the steps: employing a Weibull distribution function and an improvement factor which represents the influence of planned maintenance on an equipment failure rate, and building an equipment failure rate model which considers the aging of equipment and incomplete planned maintenance; establishing a fault rate model of the intelligent substation line protection system by using the reliability block diagram; adopting different planned maintenance periods in the stable operation period and the loss period of the line protection system; constructing an annual average operation cost model of the intelligent substation line protection system, considering the sampling trip-out modes of the direct mining direct tripping mode, the direct mining network tripping mode, the network mining network tripping mode, the single line protection configuration, the double line protection configuration mode and the maintenance cost of different maintenance types, and solving to obtain the optimal planned maintenance period of each of the two stages when the annual average operation cost is minimum. The influence of equipment aging, maintenance type, protection configuration and sampling tripping mode is considered, and the maintenance cost can be effectively reducedwhile the reliability of the protection system is improved.

The invention relates to the technical field of intelligent service pushing, and particularly discloses an intelligent service recommendation method based on a neural network mining model, which comprises the following steps: acquiring user data and preprocessing the user data; taking the preprocessed user data as data source input, and constructing a neural network model in combination with an activation function, training data and adjusting errors; and calculating a utility function of the user according to data output of the constructed neural network model, calculating user similarity anduser interestingness of the user to a service product in combination with a recommendation algorithm, constructing a hybrid service recommendation model, and establishing and displaying a hybrid service recommendation list by the constructed hybrid service recommendation model. According to the service intelligent recommendation method based on the neural network mining model, the coverage serviceproduct range is wider, the same set of model is utilized, the service recommendation standard is unified, and the priority ranking problem of service product recommendation can be solved while the accuracy and timeliness of service product recommendation are met.

The invention is applicable to the technical field of retrieval. A method for cross-modal retrieval is provided, which includes using a stacked restricted Boltzmann machine and a multi-modal depth confidence network to extract modal friendly features and modal mutual features of image and text respectively, wherein the modal friendly features can make the statistical characteristics between the obtained features more similar to the input, and modal mutual features can get the mutual information lost in the original input instances, and the two features can be fused to get the mixed features, and the final shared features can be obtained by multiple bimodal automatic coding. The embodiment of the invention utilizes a stacked restricted Boltzmann machine to extract the internal characteristics of each mode, and the mixed features suitable for cross-modal retrieval are constructed by fusing the lost mutual information between the two features in depth confidence network mining. The accuracy and retrieval speed of the cross-modal retrieval task are effectively improved by using the multi-layered and bi-modal automatic coding network mining the complex information of the cross-modal.

The invention discloses a large-scale complex network relational map mining and analysis platform and method and a compute readable storage medium, which are used to solve the problem of the large-scale application analysis caused by the multi-source heterogeneous data fusion in complex network depth mining. The invention solves the problem of analysis and mining of large-scale network from the viewpoint of complex network system, constructs a research map relation network mining analysis platform, at that same time, provides the practical application analysis and display of the complex network such as actual network key node retrieval, multi-source data fusion, network key node and hidden node mining and the like for the existing service aspect.

The invention provides a method and a system for searching whether-class problem key sentences in a reading understanding task. The method comprises the steps of selecting existing reading understanding question and answer data, preprocessing the question and answer data to obtain a data set, and then mining semantic information of questions and sentences in chapter paragraphs in the data set based on a coding layer network to obtain word embedding representation of each word; constructing an algorithm model, and calculating the questions and the text paragraphs mined through the coding layernetwork by using a neural network model and a TFIDF to obtain key sentences of whether problems are classified or not; and inputting to-be-read understanding question and answer data into the trainedalgorithm model, and predicting whether key sentences of questions are classified or not. According to the method, more key sentence supports can be provided, the weight of the key sentence is calculated through the combination of the bidirectional gating loop network and the TF-IDF, and the efficiency and accuracy of answering the whether-class problems are improved.

The invention provides a multi-disease variable site analysis platform based on a function network. The multi-disease variable site analysis platform comprises a variable gene sequencing detection module, a function enrichment analysis module, a function network construction module, a function network mining module and a shared molecular network identification module, wherein the variable gene sequencing detection module is used for finishing the fundamental analysis of sequencing data; the function enrichment analysis module is used for utilizing a function enrichment analysis tool to analyze a variable gene function; the function network construction module is used for constructing the function network according to a function enrichment analysis result; the function network mining module is used for screening stable network modules from the function network; and the shared molecular network identification module is used for identifying molecular modules shared by different diseases according to the network module. The multi-disease variable site analysis platform analyses differences of different diseases on an aspect of genomic level, establishes an incidence relationship among the diseases from the perspective of molecular function, can systematically identity a shared modular function module among different diseases, effectively analyzes the nosogenesis of similar phenotype diseases, discloses differences between the diseases from the level of genome, increases the comprehensive understanding of the diseases and is favorable for clinic diagnosis and treatment.

The invention discloses an industrial system anomaly detection method based on a graph attention network and an LSTM automatic coding model. The method comprises the following steps: 1) sample division and standardization: dividing original industrial system data into samples by adopting a sliding window; 2) building an anomaly detection model: building the anomaly detection model by adopting a graph attention network and an LSTM (Long Short Term Memory) automatic coding machine; and 3) performing real-time anomaly detection: calculating an anomaly degree score based on a reconstruction error, and performing anomaly state judgment on the basis. According to the method, the automatic coding machine is adopted, the anomaly detection model is trained in an unsupervised mode, and an anomaly annotation sample does not need to be provided; and the graph attention network is adopted to mine the association between different dimensions of the industrial system, and the anomaly detection accuracy in the complex industrial system is improved.

The invention relates a mining method of coals stored under waters, buildings or railways, which is characterized in that network mining, permanent support and integrated construction are carried out to coal strips and blocks stored under the waters, the buildings orthe railways; the network mining includes that a room, five to six meters wide from the bottom to the top is mined in a coal mine under the waters, the buildings or the railways, 10 to 20m posts are left and the room and the posts form network shape; the permanent support includes that the mined room is treated with the permanent support in time, and the permanent support has large intensity to guarantee that the deformation of the top plate and the two sides is comparatively small and ground above the waters, or of the buildings or the railways does not sink; the integrated construction includes that the technologies of hole drilling, explosion, coal loading and transport are adopted and equipment with integrated support is used for carrying out construction to the mined room. The mining method of the invention has the advantages that resource recovery rate is improved, investment is low and return rate is high; and not only an underground treatment space is provided for the recrements of a mine well, but also filling materials are provided for the mining under the waters, the buildings or the railways, thereby realizing the advantages of safe production, etc.

The invention provides a sequence recommendation method for mining long and short term interests of a user based on a graph neural network, which comprises the following steps of: obtaining personal information of the user and a user interaction sequence data set, preprocessing the data set and dividing the data set into a training set and a test set; constructing a sequence recommendation model for mining long and short term interests of the user based on a graph neural network; training the sequence recommendation model for mining the long and short term interests of the user based on the graph neural network; inputting the personal information and the interaction sequence of the to-be-recommended user into the trained sequence recommendation model for mining the long and short term interests of the user based on the graph neural network, calculating the recommendation score of the to-be-recommended item relative to the user, and recommending the item to the user according to the recommendation score; the method solves the problems that long-term and short-term interests of a user cannot be effectively captured in a sequence recommendation scene, and noise is difficult to distinguish.

The invention relates to a terrorist organization network mining algorithm which comprises the following steps of: S1, constructing a suspected node connection network; S2, calculating a threat degree measured value of each node in the suspected node connection network; S3, calculating a topology potential of each node in the suspected node connection network; S4, sorting the topology potential of each node, which is obtained in the step S3, by adopting a rapid sorting method, and finding out a local maximum potential value node; and S5, by using the nodes with the relatively high topology potentials as center nodes, outputting terrorist organization networks N1, N2,..., Nt. Compared with a conventional method for judging node importance of the terrorist organization networks by using a node degree as an index, the method can more efficiently mine the terrorist organization networks and important nodes in the terrorist organization networks and discloses an internal network structure among terrorist organizations.

This invention relates to building a landmark database from web data. In one embodiment, a computer-implemented method builds a landmark database. Web data including a web page is received from one or more websites via one or more networks. The web data is interpreted using at least one processor to determine landmark data describing a landmark. At least a portion of the landmark data identifies a landmark. Finally, a visual model is generated using the landmark data. A computing device is able to recognize the landmark in an image based on the visual model.

The invention discloses a short video click rate prediction method based on a sequence capsule network. According to the method, based on a click sequence of a user to a short video, multiple interests of the user are mined by utilizing a sequence capsule network, and the click rate of the user to a target short video is predicted. The method is mainly composed of three parts: in the first part, context features are extracted from a user click sequence by using a convolutional neural network; in the second part, a sequence capsule network is used for converting the context features into different interest spaces, the seriousness of user behaviors is captured in the different interest spaces, and multi-interest vector representation of the user is obtained; and in the third part, the clickrate of the short video is predicted based on the multi-interest vector representation of the user.

The invention discloses an Internet of Things data mining method, the Internet of Things data mining method comprises the steps of classification, review analysis, clustering, association rules, features, change and deviation analysis and Web page mining, and the operation mode of the method is divided into the following specific steps. According to the Internet of Things data mining method, afterthe Internet of Things is mined in various modes, the mined data can be analyzed and processed, one or more superior operation modes are selected according to the data mining information during processing, and the analysis mode is verified multiple times through the corresponding steps, Then, the data with relatively stable data is selected as evaluation preparation data, and the data is input into a computer for virtual simulation calculation and analysis of the operation mode, so that the data tends to be accurate through multiple times of verification and analysis of the network mining data, and the network data is subjected to corresponding computer simulation; and the operation risk can be avoided to a greater extent.

The invention discloses a microblog topic mining method based on heterogeneous social media user dynamic behaviors, which comprises the following steps: constructing an attribute multivariate heterogeneous dialogue network, and mining heterogeneous social context for topic detection; introducing a neighbor-level attention mechanism and an interaction-level attention mechanism to model different influences of different neighbors and different types of interaction modes on topic inference, and learning embedding of a specific view; the representations of the plurality of views are used as inputsof multi-view neural variational reasoning, and complex associations among topic semantics carried by different views are captured, so that topics with better consistency are mined.

The invention provides an interaction behavior prediction method based on sequential network mining, which is comprehensive in communication interaction behavior rule summarization and capable of accurately predicting interaction behaviors, and comprises the following steps: constructing an interaction behavior sequential network according to network communication interaction behavior records; according to a period target parameter and a season target parameter, screening period and season sub-graphs from the time sequence network as nodes, and constructing a sub-graph spanning tree; and according to an attention target parameter, screening out a maximum period season sub-graph from the sub-graph spanning tree, determining a network communication interaction behavior rule, and predicting a network interaction behavior by using the network communication interaction behavior rule. The device comprises a sequential network module, a parameter setting module, a sub-graph spanning tree module, a sub-graph screening module and a behavior prediction module. The electronic equipment comprises a memory, a processor and a computer program which is stored on the memory and can run on the processor to realize the interactive behavior prediction method based on sequential network mining.

The invention relates to a method for mining comparable network language materials. The method includes acquiring source language web pages by the aid of network crawlers and preprocessing the source language web pages to obtain source language documents; analyzing probabilities of cross-language topics of the source language documents and generating corresponding target language query phrases; submitting the target language query phrases to search engines and selecting front N documents to form a target language candidate similar document set; computing similarity degrees of the source language documents and target language candidate similar documents, sieving documents with high similarity degrees and constructing a comparable language material bank. The invention further discloses a device for implementing the method for mining the comparable network language materials. The method and the device have the advantages that the problem of ambiguity or long time consumption due to vocabulary translation can be solved; the source language documents come from specific website contents acquired by the network crawlers, the target language documents come from the integral internet, and accordingly the source language document utilization rate can be effectively increased; the source language documents are matched with the target language similar documents by the aid of topic distribution similarity, and accordingly the language material bank construction accuracy can be improved.

The invention provides a coordinate inversion method and system. The coordinate inversion method comprises the following steps: S1, receiving global positioning system (GPS) data of a spatial position and network mining data, comparing the attributes of the GPS data and the network mining data, and finding out the data with the same attribute; S2, confirming a plurality of sample points in the data with the same attribute, and calculating transformation parameters according to the plurality of sample points; and S3 carrying out coordinate inversion on the network mining data according to the transformation parameters. Compared with the prior art, the coordinate inversion method and system provided by the invention are high in degree of automation and processing efficiency.

The invention relates to a robust training method for a target detection network, and the method comprises the steps: obtaining a training sample, wherein a part of detection targets on the training sample carries a manual labeling box; performing feature extraction on the training sample by using a target detection network, and generating a suggestion box on the training sample; marking an original sampling label on the suggestion box, wherein the original sampling label comprises a positive label and a negative label; carrying out pooling operation on the positive label by adopting a poolingbranch, and outputting a first region-of-interest feature; inputting the first region-of-interest feature into a mining network which is a fully connected neural network, and generating a new suggestion box label, namely a mining label, by the mining network; fusing the mining label and the original sampling label to generate a gold label; and applying the gold label to the training of the targetdetection network.

The invention discloses an image grading method, device and equipment and a storage medium. The method comprises the steps: determining an original three-dimensional image corresponding to an originalAS-OCT image; sequentially inputting the intermediate three-dimensional images of the first preset number scale corresponding to the original three-dimensional image into a corresponding preset 3D convolutional neural network to obtain a corresponding one-dimensional vector; performing calculating according to the first preset number of one-dimensional vectors to obtain a corresponding output result; and determining the turbidity degree of the original AS-OCT image according to the output result and a pre-configured turbidity category. According to the invention, the original AS-OCT image isshot from different angles, more features in the image can be extracted and learned, and the network classification precision is effectively improved; meanwhile, by constructing a multi-scale 3D convolutional neural network, intermediate three-dimensional images of multiple scales corresponding to an original three-dimensional image are input into a corresponding preset 3D convolutional neural network, so global features and local features are fused to facilitate network mining to obtain more discriminative feature information.

The invention discloses an alarm root cause identification method based on causal network mining and a graph attention network, and solves the problem of rapid and accurate fault positioning of a large-scale complex communication network. Starting from the reality of network equipment alarms, a maximum and minimum hill climbing method (MMHC) is used for mining causal trigger relationships among the alarms, and on the basis, a graph attention network is used for accurately positioning the alarms. The model has certain fault tolerance for the mined alarm relationship, and the weight influence ofdifferent neighbor nodes is adjusted through an Attention mechanism, so that the identification of the root cause alarms is more accurate, and 93% of identification accuracy is achieved.

The invention relates to an airplane automatic driving operation simulation method based on a long-short term memory network, and belongs to the field of airplane automatic driving. The whole-process flight data of the air route is used as a training set, the correlation of the data in the time sequence is mined by using a long-short-term memory network, and the mode that a pilot makes a driving behavior decision according to the navigation information of the air route is learned. Through training, a model learns key decision information of flight mode conversion performed by a human pilot according to navigation data. Flight mechanism analysis and data correlation analysis are carried out on independent flight stages, and corresponding model training input is determined. Through training, the model learns a mapping relation from input of a flight state, a flight environment and the like to output of operation variables. Therefore, in the actual flight process of an aircraft, according to the sensed flight state, flight environment and other data, the corresponding operation variables of the throttle lever, the pedal and the pitching rolling rocker are obtained through prediction of the long-short-term memory network model, and therefore automatic driving of the aircraft is achieved.

The invention provides a target detection method. The method comprises the following steps: extracting image features to generate a feature map; performing up-sampling on the feature map to obtain an amplified feature map; connecting the amplified feature map to a category prediction head, a width and height prediction head and a central point offset prediction head; adding a category attention network into the category prediction head, and mining effective information between targets which are far away from each other within the category and between the categories but are semantically related; supervising training of each prediction head through supervising information generated by encoding a real target frame; and frame-selecting an identification object in the image to be detected according to a result output by each prediction head, and marking a classification result. According to the method, category attention for further judgment of target categories and scale adaptive coding for frame regression are combined, so that the network can associate intra-class and inter-class features, and effective information between intra-class and inter-class targets which are far away from each other and are semantically related is mined, meanwhile, more accurate frame selection can be carried out according to the scale change of the detection target, so that the detection accuracy and the frame selection precision are improved.

The invention discloses a super-resolution reconstruction method based on deep learning local and non-local information. The method mainly comprises the following steps: establishing a super-resolution convolutional neural network model based on deep learning local and non-local information, wherein the super-resolution convolutional neural network model comprises a local network module and a non-local enhancement network module; respectively training super-resolution models of different amplification factors by using the convolutional neural network built in the previous step; and taking thetrained super-resolution reconstruction model as a basis, taking the low-resolution image as input, and obtaining a final super-resolution reconstruction image. According to the method provided by theinvention, effective information of a wider area of the image can be mined by utilizing the non-local enhancement network, so the super-resolution reconstruction can be effectively carried out on thelow-resolution image, a good subjective and objective effect can be obtained, and the method is an effective low-resolution image restoration method.