48 results about "Interaction nets" patented technology

The invention provides an abnormal account identification method and system. Through construction of a data interaction network, network attribute features of the data interaction network and non-network attribute features of the data interaction network are obtained, then according to the network attribute features of the data interaction network and the non-network attribute features of the data interaction network, a classification prediction model is obtained, and an account to be identified is identified by use of the classification prediction model, such that an abnormal account does not have to be identified through manual discrimination, numerous manpower resources are saved, and the identification efficiency is effectively improved. At the same time, by use of a scientific digital analysis means, the identification accuracy is improved.

The invention relates to a method for discovering topics of communities in an on-line social network. The method includes the specific steps that data acquisition is carried out on the object social network based on a web crawler; the relevancy of each user object in an interactive network topological structure is worked out based on an acquired interactive relationship between the user objects in the social network; a static interactive network of the user objects is constructed; a compact user community structure is obtained through hierarchical clustering according to the relevancy of each user object; for each community obtained through division, a database is searched to acquire text messages corresponding to the community, the text messages are input as documents and classified through an SVM, and the hot topics of the community are worked out. Compared with an existing method for discovering topics in an on-line social network, the method for discovering the topics of the communities based on community division has the advantages that noise data can be effectively removed, the more compact topics between the communities can be obtained, and a deeper understanding of information spreading laws of the social networks is facilitated.

The invention discloses a method and device for determining an abnormal interactive account. The method comprises the steps that interactive data among a plurality of accounts is obtained; based on the interactive data, an interactive network of the accounts and at least one key account of the interactive network are determined; at least one interactive group containing the key account in the interactive network is determined; based on feature information of the interactive group, it is determined whether or not the interactive group is an abnormal interactive group, and the abnormal interactive group contains an abnormal interactive account. The method and device for determining the abnormal interactive account have the advantages that the technical problem in the prior art is solved thataccount containing abnormal interaction is determined just by the amount of the interactive content among accounts so that the accuracy of the result is lower.

The invention discloses a lightweight fine-grained image recognition method for cross-layer feature interaction in a weak supervision scene, and the method comprises the steps: constructing a novel residual module through employing multi-layer aggregation grouping convolution to replace conventional convolution, and enabling the novel residual module to be directly embedded into a deep residual network frame, thereby achieving the lightweight of a basic network; then, performing modeling on the interaction between the features by calculating efficient low-rank approximate polynomial kernel pooling, compressing the feature description vector dimension, reducing the storage occupation and calculation cost of a classification full-connection layer, meanwhile, the pooling scheme enables the linear classifier to have the discrimination capability equivalent to that of a high-order polynomial kernel classifier, and the recognition precision is remarkably improved; and finally, using a cross-layer feature interaction network framework to combine the feature diversity, the feature learning and expression ability is enhanced, and the overfitting risk is reduced. The comprehensive performance of the lightweight fine-grained image recognition method based on cross-layer feature interaction in the weak supervision scene in the three aspects of recognition accuracy, calculation complexity and technical feasibility is at the current leading level.

A recommendation method based on the graph interaction network is applied to the field of user personalized recommendation. A traditional recommendation method and a traditional deep learning method are difficult to meet complex application environments due to rapid development of the Internet industry and continuous increase of the network data volume, and have defects in the aspects of accuracyand space complexity. Therefore, the invention provides the recommendation method based on the graph interaction network, by adopting the method, the personalized recommendation accuracy can be ensured, the space complexity of the model is reduced, and the method has a wide application prospect.

The invention discloses a cross-modal time sequence behavior positioning method and device for a multi-granularity cascade interactive network, and aims at solving the problem of time sequence behavior positioning based on given text query in an unpruned video. According to the invention, a new multi-granularity cascade cross-modal interaction network is implemented, cascade cross-modal interaction is carried out in a coarse-to-fine mode, and the cross-modal alignment capability of the model is improved. In addition, the invention introduces a local-global context-aware video encoder, and the local-global context-aware video encoder is used for improving the context time sequence dependence modeling capability of the video encoder. The visual-language cross-modal alignment method is simple in implementation method and flexible in means, and has the advantage of improving visual-language cross-modal alignment precision, and the model obtained through training can remarkably improve time sequence positioning accuracy on paired video-query test data.

The invention discloses a complex simulation system credibility evaluation method based on a network topology path, and belongs to the field of system simulation. And under the condition that the nodecredibility of the single model is known, the credibility of the whole complex simulation system is quantified. The method comprises the following steps: firstly, analyzing an information interactionrelationship between component models in the complex simulation system, calculating weights of edges between nodes according to objective indexes, and abstracting the complex simulation system into adirected weighted model interaction network; and calculating the out-degree of each node in a model interaction network, selecting a node with a relatively high out-degree as an initial node, starting from the initial node, obtaining different single execution paths, calculating the credibility of the single execution path, and integrating the credibility of all the execution paths to obtain thecredibility of a simulation system. Aiming at a complex equipment simulation system with the characteristics of complex mechanism, complex input and output variables, strong uncertainty and the like,the invention solves the problem that it is difficult to carry out quantitative analysis on evaluation of the credibility of the complex equipment simulation system.

The invention discloses a personalized recommendation method and system based on context awareness and feature interaction modeling. The invention belongs to the technical field of data mining, in order to solve the technical problem of how to recommend for a user according to preferences of the user in different context environments and improve the accuracy of recommendation, the adopted technical scheme is as follows: a feature interaction network model based on context-aware feature interaction is constructed, and the method specifically comprises the following steps: constructing a contextfeature information attribute model; constructing a context feature information-user / context feature information-article interaction model; constructing an influence degree model of different contextfeature information on the user / article; constructing an overall influence model of the context environment on the potential feature information of the user / article; and constructing a feature interaction network prediction model. The invention further discloses a personalized recommendation system based on context awareness and feature interaction modeling.

Methods and devices for clustering a plurality of sub-networks of a larger interaction network using an enhanced hierarchical clustering algorithm are disclosed. The methods provide expression based sub-network generation using differentially expressed markers. The enhanced hierarchical clustering algorithm clusters the generated sub-networks based on a user defined customizable similarity coefficient. The methods use non-Boolean links to cluster similar sub-networks. This provides consideration of indirect relationships among sub-networks. The customizable similarity coefficient enables the methods to be used for diverse applications such as biomarker detection, patient stratification, personalized therapy, drug efficacy prediction, genetic similarity analysis in genetic diseases. The methods enable patient grouping based on the enhanced hierarchical clustering algorithm.

The invention discloses a recommendation system score prediction method based on a graph neural network and an attention mechanism, and the method comprises the following steps: S1, converting a userproject score graph into a user project score credibility graph based on degree and time information, and sampling neighbor vertexes for each vertex in the graph; s2, updating the state of each vertexin the user project scoring credibility graph by utilizing a sampling result and combining time information; s3, performing score prediction on the project by the user based on an attention mechanism, and updating a score prediction model; and S4, using the score prediction model to realize score prediction of the project by the user. According to the invention, an application way of the graph neural network in the universal recommendation system is given, static characteristics of users and projects are combined, the graph representation learning ability of the graph neural network is utilized to learn the importance degree of hidden characteristics in the user project interaction network, and more attention is paid to serve the recommendation system.

A method and system constructs a socio-technical network representing design and development processes. In one aspect, a network of inter-personal interactions comprising at least a plurality of nodes representing actors in design and development process is established; an artifacts network comprising at least a plurality of nodes representing a plurality of heterogeneous artifact types is established; one or more relationships between the nodes in the network of inter-personal interactions are determined; one or more relationships between the nodes of the artifacts network are determined; and one or more relationships between the nodes in the network of inter-personal interactions and the nodes of the artifacts network are determined.

In a method and apparatus for analyzing multivariable data sets, a general computerized platform is provided for evaluating the relationship between large number of measurements of sets of variables characterizing components of complex states of a system under induced stimulation or controlled conditions. The linked responses of variables and their temporal relations tell about the network of interactions and their hierarchy. Processing of data sets by a simple neural network gives a matrix of weight parameters, that allow to identify fingerprints of complex states characterized by patterns of measured variable and estimate the interactions between the components characterized by the measured variables. The results are provided numerically and by color-coded presentation indicating dominating relations between variables and strongly responding variables. When applied to dynamic responses of a system, the analysis can construct a schematic hierarchical architecture of the network of interaction between the components of the studied system. Applications in biology include analysis of measurements characterizing responses of molecular components in cells under changes induced by stimuli (e.g. drugs, growth factors, hormones, mutations or forced expression of a proteins), and identification of complex cellular states (e.g. proliferation, differentiation, transformation, starvation, necrosis, apoptosis, and the time dependencies of the above effects).

The invention discloses an emotion instability user detection method for video conventional comments, which comprises the following steps of: 1, collecting original data from a website, and screeningout a conventional comment text, user information and the comment time; 2, achieving formatting processing of the data to form a conventional comment set; 3, summarizing preset expression packages ofthe website, constructing an expression comparison table, and measuring emotional tendency of conventional comments to achieve emotional analysis of the conventional comments. 4, constructing a time sequence comment interaction network; and step 5, judging whether the user conforms to the definition of the user with unstable emotion or not according to the time sequence inter-user relationship, soas to detect the user with unstable emotion. The method has the advantages that emotion-variable users are found according to conventional comment content in the video in the early stage of public opinion development, great convenience is brought to control and guidance of public opinions, and good social benefits and economic values are generated.

The invention relates to the technical field of cloud computing, and aiming to solve the task scheduling problem of a cross-regional data center and optimize the scheduling result of a task in the data center so as to minimize the overall completion time, the invention relates to a cross-data center task scheduling and bandwidth allocation method based on hypergraph segmentation. The method is composed of a task scheduling control part and a network bandwidth distribution part, the task scheduling control part is responsible for realizing scheduling and deploying the tasks among the data centers based on network interconnection, periodically detects the arrival of a job and starts a scheduling algorithm, and generates a scheduling scheme by combining the information, such as a data centerresource capacity and a task completion state which are acquired by current monitoring, etc.; and the network flow control part is divided into two types of a real-time online application interactivenetwork flow and a data transmission non-interactive network flow in a task scheduling process. The method is mainly applied to the communication control occasions.

The present invention relates to systems and methods for identifying a biomarker from associative and knowledge based systems and processes. Particularly, aspects of the present invention are directed to a computer implemented method that includes data mining one or more public sources of biomedical text, scientific abstract, or bioinformatic data using queries to identify database terms associated with one or more predetermined terms, scoring association(s) between each of the identified database terms and the one or more predetermined terms, determining a subset b based on the score of the association(s), developing an interaction network model comprising the database terms in subset b, interactions, and additional database terms using a combination of algorithms in a predetermined order, and identifying candidate biomarkers from the interaction network model based on a ranking of the database terms in subset b and the additional database terms in the interaction network model.

The invention discloses an information diffusion prediction method based on a time sequence hypergraph attention neural network. Information diffusion is predicted by learning preference of a user from two aspects of a static friendship network and a dynamic interaction network of the user. According to the method, not only is a static dependency relationship of a user captured from a friendship network of the user by using a graph convolutional neural network, but also a hypergraph attention network is innovatively designed, so that interaction of the user on a cascade level and connection between cascades are dynamically learned from a serialized information diffusion hypergraph. And according to the cascade characteristics to be predicted, an embedded search module searches vectors of corresponding users from the obtained user representation vectors in the two aspects so as to carry out the next step of interactive learning. And finally, two self-attention modules are utilized to respectively carry out internal deep interactive learning on the cascade representation obtained from the two aspects to predict the next influenced user, so that gradual prediction of network information diffusion is realized.

Embodiments disclose a system for determining a sensitivity of a networked system. The system identifies a vertex (V) that represents a constituent and a state of the constituent, and an interaction (E) between at least two Vs. An interaction-dependent function provides a probability that, when a perturbation occurs, a V will be in a certain state given that it is currently in a determined state and its neighbors are currently in determined states. A network reliability is used to determine a probability that a V's state holds when a perturbation occurs. The system evaluates only a certain amount of terms in a Taylor series from a sample, and identifies interpolating polynomials between the Taylor series. A cost function optimizes a property of the networked system for a fixed cost. The system perturbs the networked system until reliability is zero to estimate a sensitivity of the networked system.

In recent years, a recommendation method based on a graph neural network achieves great success in academic and industrial circles, some research scholars simulate the social influence of recursive propagation in a social network through a high-order relationship among graph convolutional neural network modeling users, and feature vectors of high-order neighbors are utilized to constrain feature vectors of target users. In order to improve the accuracy of social recommendation, the influence propagation of the collaborative similarity between the user and the article hidden in the user and article interaction network is further captured, and the preference of the user changes along with the propagation of the social influence and the collaborative similarity influence. In combination with different characteristics of information representation of an actual recommendation scene, a user social domain and a user article interaction domain, the invention adaptively initializes a user potential feature vector in different semantic spaces to reflect the characteristic that a social relationship between users and an interaction relationship between user articles generate different influences on constraint user feature vectors. In addition, in order to enable the model to be more suitable for practical application, the invention discloses a rapid non-sampling optimizer to learn model parameters, and the model optimization efficiency is improved.

The invention provides a knowledge graph question and answer method and system based on a neighbor interaction network. The method comprises the following steps: obtaining a knowledge graph, converting the knowledge graph into embedded expressions of entities and relationships of the knowledge graph according to the knowledge graph and a neighbor interaction network, and forming a semantic space; representing a question according to the question and a pre-training language model to obtain a vector representation of the question; putting the vector representation of the question into the semantic space to predict an answer entity to obtain an answer to the question.

The invention provides a fine-grained article recommendation method and system based on comment texts. The method comprises the following steps: respectively obtaining a user comment text set and an article comment text set; calculating a multi-granularity incidence matrix between the user and the article by using a fine-granularity feature interaction network to obtain a 3D interaction image; inputting the 3D interaction image into a full-connection neural network and a traditional factorization machine to realize user-article score prediction, and realizing article recommendation according to a score result. According to the scheme, user and article comments are coded through a multi-level expansion convolution structure, loss of fine-grained information in the comments is avoided, feature interaction of the user and article comments is constructed under multiple granularities, and multi-granularity information is fused and processed through 3D convolution, so that related information under the multiple granularities in the comments is effectively highlighted, and the rationality and accuracy of article recommendation are effectively improved.

The invention belongs to the technical field of block chains, and particularly relates to a material assistance traceability system based on Fabric. The system provided by the invention comprises a client, an IPFS (Internet Protocol File System), a Web service, a FabricSDK (Software Development Kit) and a Fabric network. The client provides an operation use interface for the user; the Web serviceprocesses the request received by the IPFS from the client; the Fabric network is composed of an account book, a channel, a chain code, a CA, a peer node and a sorting node. The CA can manage the authority and resources of the role in the Fabric network; wherein the chain code is used for realizing main service logic of the system and performing various interactions with an account book; internalnodes of the Fabric network are provided with interfaces based on a gRPC protocol and are used for data interaction; the Fabric also provides an SDK (Software Development Kit) of multiple language versions; various resources in the Fabric network, including account books, transactions, chain codes, events and authority management, can be accessed through the SDK. The system can publish demands andassistance, record assistance material transfer information and ensure authenticity and traceability.

In a method and apparatus for analyzing multivariable data sets, a general computerized platform is provided for evaluating the relationship between large number of measurements of sets of variables characterizing components of complex states of a system under induced stimulation or controlled conditions. The linked responses of variables and their temporal relations tell about the network of interactions and their hierarchy. Processing of data sets by a simple neural network gives a matrix of weight parameters, that allow to identify fingerprints of complex states characterized by patterns of measured variable and estimate the interactions between the components characterized by the measured variables. The results are provided numerically and by color-coded presentation indicating dominating relations between variables and strongly responding variables. When applied to dynamic responses of a system, the analysis can construct a schematic hierarchical architecture of the network of interaction between the components of the studied system. Applications in biology include analysis of measurements characterizing responses of molecular components in cells under changes induced by stimuli (e.g. drugs, growth factors, hormones, mutations or forced expression of a proteins), and identification of complex cellular states (e.g. proliferation, differentiation, transformation, starvation, necrosis, apoptosis, and the time dependencies of the above effects).

The invention discloses a knowledge-based retrieval dialogue method based on pre-training and an attention interaction network, and the method comprises the following steps: training a pre-training language model BERT on a target corpus through employing a domain adaptability pre-training method, and obtaining the domain adaptability BERT; using the field adaptability BERT as an encoder of the attention interaction network, and respectively encoding the dialogue context, the background knowledge and the plurality of candidate response texts to obtain corresponding representations; and finally, respectively inputting the dialogue context, the background knowledge and the representation of the plurality of candidate responses into the attention interaction network for matching, and training the attention interaction network to retrieve the optimal response from the plurality of candidate responses. According to the method, the powerful semantic characterization capability of the pre-training language model is utilized, the semantic characterization capability of the pre-training language model on a specific corpus is improved through two pre-training tasks, and the performance reduction caused by separation coding adopted for improving the retrieval speed is relieved by adopting the attention interaction network.

The invention discloses an automatic question and answer generation method and system, a terminal and a readable storage medium, and overcomes the defect of insufficient coding capability of an incomplete knowledge graph in the prior art through a global normalized graph attention network, a coarse and fine granularity combined rich semantic text reading network and a problem and entity relationship deep interaction network, rich semantic text information is insufficient, and deep interaction between problems and entity relations is lacked. The accuracy of the question and answer result is improved; meanwhile, through a text content reading module combining coarse and fine granularities, on the basis of utilizing entities in the text and related information, relationship characteristics among the entities implied in the text can be further mined to complete the knowledge graph. And finally, a bidirectional attention network is utilized to carry out deep information interaction on the user question and the entity so as to find the entity more related to the user question.

The invention discloses a click rate estimation method based on a multi-domain partition integrated network, which is characterized by adopting three parallel modules, providing two multi-domain partition strategies for original onehot form features and dividing into region vectors; independently embedding each area vector by using a segmentation embedding method to obtain an embedded layer vector; and sharing the embedded vector to a mining middle-order interactive network and a high-order interactive neural network. The middle-order interactive network adopts the FFM to extract the feature interaction between the regions, and the high-order interactive part introduces the integration thought on the basis of the neural network, so that the degree of parallelism of the network is widened, and the expression ability of positive features is enhanced. According to the method, independent features and interactive features are considered at the same time, the expression ability of an embedded layer is enriched by using the idea of segmented embedding, the neural network is extended under the condition that space complexity is not introduced, the problem of gradient disappearance is effectively solved, and the click estimation ability is improved.

The invention relates to the technical field of data processing, in particular to a prediction method and device for a service interaction network, and solves the problems that an established service interaction network is incomplete and a potential service relation cannot be predicted. The method comprises the steps of obtaining collected service data, establishing an interaction network, establishing an adjacent matrix and a feature matrix, and obtaining a prediction result; and inputting the adjacent matrix and the feature matrix into a prediction model to obtain a service interaction prediction matrix, processing elements in the service interaction prediction matrix to generate a service interaction network matrix, and determining a predicted service interaction network based on the service interaction network matrix. Therefore, the interaction network and the prediction service interaction network generated by adopting the service data avoid the defect that the established service interaction network is incomplete due to the fact that all interaction objects of the same service type need to be retrieved and determined when the prediction interaction network is established, and the interaction objects with potential interaction possibility can be predicted.

The invention relates to a user abnormal transaction account detection method based on a mahalanobis distance technology. Compared with the prior art, the defect that abnormal transaction account detection is difficult to meet actual use requirements is overcome. The method comprises the following steps: acquiring user account activity data; constructing an interactive network directed graph; an egonet model is initialized; acquiring data of a to-be-detected user account; feature extraction of the egonet model is carried out; and detecting the abnormal transaction account of the user. According to the method, the transaction activity of the user account under the transaction network model is depicted and analyzed to design the method for detecting the abnormal transaction account of the user based on the structured graphic data, so that the performance of a detection algorithm is improved, and the omission of the abnormal activity is reduced.

The invention discloses a text feature extraction method based on machine learning, and the method comprises the following steps: 1, initializing a system; 2, inputting data; 3, carrying out part-of-speech tagging; 4, training a machine learning block model; 5, carrying out text partitioning; and 6, carrying out text output; wherein in the step 1), a display is arranged on the outer wall of one side of the SVO block text extractor, in the step 2), data of external equipment can be input into the SVO block text extractor through an interactive network module, and in the step 5), SVO block texts are semantically related mark groups.According to the method, a plurality of part-of-speech marks are identified in an unstructured text; a plurality of SVO block texts are determined from a plurality of part-of-speech tags by using a machine learning block model, and the machine learning block model is trained on training data marked with a subject-verb-object (SVO), so that the time cost of text feature extraction is reduced, and manual development and rule updating are not needed.