Group discovery method based on data enhancement and non-negative matrix sparse decomposition

Abstract

The invention belongs to the technical field of big data, and particularly relates to a group discovery method based on data enhancement and non-negative matrix sparse decomposition. According to themethod, neighbor distribution of nodes with different attributes in a network with attributes of the nodes is learned through the recurrent neural network; a neighbor mode formed by similar neighbor distribution of a plurality of nodes is extracted to enhance data, a non-negative matrix is subjected to sparse decomposition through alternating least squares and Glynov regularization to find a group, and the group discovery accuracy is improved. According to the method, the edges of the network topology structure are filled through data enhancement, so that all-zero lines are not iterated as much as possible during matrix decomposition, and iteration stability is ensured; moreover, sparse group representation is obtained through a non-negative matrix sparse decomposition method, a large number of non-zero probabilities cannot be obtained, the interpretability of the group discovery method is enhanced, and the problem that a large number of non-zero probabilities are difficult to interpret group membership relations is solved.

Description

technical field [0001] The invention belongs to the technical field of big data, and in particular relates to a network data enhancement based on deep learning and a group discovery method of non-negative matrix sparse decomposition. Background technique [0002] In reality, there are connections and interactions between various objects, and these objects and the connections between them can be abstracted into a network structure, or called a graph structure. Connections or interactions are represented by edges in the network, while these objects and their properties are represented as nodes and their node properties in the network. Analyzing these network data to find out the set of similar points among them is called group discovery task. Group discovery is an important task in big data mining. For example, in a social network, network nodes represent everyone, while edges in the network represent the social relationships that exist between them, such as classmates, relat...

AI Technical Summary

Problems solved by technology

Different from the non-sparse decomposition method, the probability that some nodes may belong to multiple groups is similar, which makes it difficult to judge the relationship between the nodes belonging to the group; on the overlapping group discovery task, the sparse decomposition method can make each node belong to different groups. The probability of the subordinate group is close to 0, instead of the sparse decomposition method, each group may get a non-zero probability, making it difficult to distinguish which groups the node belongs to

Images