Community Detection on Social Networks With Sentimental Interaction

Community Detection Based on Semantics

Many traditional community detection methods ignore node content attributes in favor of network topology information alone. Nevertheless, node content is crucial and helpful for community discovery, because it divides the community in a way that is more in line with its actual circumstances and makes a bigger contribution to the community. Node characteristics are based on the inherent structure of individuals, which is another way to express the nature of the community.

The topological information and content information complement each other, thus producing a more accurate result. Another advantage is that even if a single source of information is lost, a relatively stable community structure can still be built with another source of information, e.g., content analysis and personality characteristics. The idea of using a topic model-based method to unify the model and then estimating each parameter to find the community distribution was proposed by Balasubramanyan and Cohen (2011) and Yang et al. (2009). Zhu et al. (2013) combined classic ideas in topic modeling with a variant of the mixed-membership block model recently developed in the statistical physics community; their model can also be used for detecting generalized communities. Block-LDA (Balasubramanyan & Cohen, 2011) combines aspects of mixed membership stochastic block models and topic models to improve entity link modeling by jointly modeling links and text about the entities that are linked. Yang et al. (2009) proposed a discriminative methodology for detecting communities by integrating link and content analysis. For content analysis, they also developed a conditional model and a discriminative model. Neville et al. (2003) proposed a weighted adjacency matrix to allow a content similarity measure. The weight of each edge was defined as the number of attribute values shared by the two end vertices. They then applied three existing graph clustering algorithms to the weighted adjacency matrix to perform network clustering. Furthermore, since semantic structures are related to each other, TCCD (Wang et al., 2019) can learn the community structure and semantic interpretation of each community and search for the correlations of different topics in community detection.