| Summary: | 碩士 === 樹德科技大學 === 資訊工程系碩士班 === 107 === Due to the increasingly faster technology and network transmission speeds, user interaction on the social media platforms implies massive and complex high-level interactions. Although previous research has developed models and tools to analyze human interaction and community patterns, most of them focus on binary relationships. Therefore, the study of multi-relationships, which comes closer to reality, takes on more importance.
This paper proposes to use Hypergraph’s hyperedge to present multi-relationships. The relationships of the hyperedge''s internal nodes were projected onto the binary relationship diagram with weighted distance. With Degree Centrality determining the center of the overlapping community, a bottom-up cluster was formed using the degree of association between nodes and community, the weighted distance between nodes, as well as local clustering coefficients. In the experiment, the input Hypergraph first established the number of hyperedge nodes by modified versions of the long tail theory and normal distribution. Then nodes within the hyperedge were randomly generated. The randomly established Hypergraph was investigated, the clustering threshold setting was determined, performance evaluated, and the result was compared with the CPM method. The results showed that, as the DcDL clustering method could adjust thresholds, it had great flexibility than the CPM in overlapping community discovery.
|
|---|
