Implicit group discovery method based on latent features of online social users

Abstract

The invention provides an implicit group discovery method based on hidden features of online social users, and relates to the field of network technology. The method includes the following steps: constructing user social relationship matrix for social network; using sparse self-encoder to learn hidden features of user social relationship; using hidden layer output encoding matrix for robust continuous clustering; using feature matrix to construct graph structure, from graph structure Identify implicit groups in the connected branches of . By discovering the internal mechanism of user connection, the present invention considers the connection relationship between users from the perspective of hidden features, which is more in line with the real hidden group aggregation situation, and realizes more accurate discovery of hidden user groups.

Description

technical field [0001] The invention relates to the technical field of social networks, in particular to an implicit group discovery method based on latent features of online social users. Background technique [0002] With the development of Web2.0 applications and other types of social media, online social networks (onlinesocial networks, OSN) have become the most important platform for people's online life. On these platforms, users do not exist alone, they may Certain recessive groups will be formed due to social choices or social influences. With the rapid development of e-commerce, the phenomenon of information overload is becoming more and more serious. As an effective tool to alleviate information overload, the recommendation system has become the standard configuration of modern e-commerce websites and social platforms. Therefore, it is possible to effectively capture the implicit group. Compared with studying the influence of the entire social network users on a c...

AI Technical Summary

Problems solved by technology

[0004] Clustering algorithms are widely used in the research of group discovery. The existing clustering algorithms include K-Means, DBSCAN, WAVE-CLUSTER, FCM, COD, GMM, spectral clustering, etc., but no clustering algorithm can be universally applicable. In order to reveal the various structures presented by various multi-dimensional data sets, the existing clustering algorithms have limited performance in the effectiveness of high-dimensional data, and many parameters must be adjusted for different research fields or data sets.

Robust continuous clustering (RCC) can be used for unsupervised clustering of large-scale data sets, suitable for different types of data, such as text, images, numbers, etc. Good performance, but this method cannot solve the problem of user clustering with only social network user connections

Images