In this paper we present an alternative way of analysing the temporal network of Reuters' news coverage of the terrorist attacks on September 11th 2001, in four selected time slices. Each time slice forms a subnetwork in which a vertex represents a word or a term that appeared at least once in the news articles. There is an edge between two words if they appear together in at least one linguistic unit or sentence and the value on the edge represents the number of these co-occurrences. We used Pajek to identify central words in all four subnetworks based on their degree, weighted degree and according to their eigenvector value. Based on the weights or values on the edges, we found the most frequent combinations of words that appeared in each sentence of the reports and visualised the results. We then used islands, the Louvain method and VOS Clustering, different techniques for detecting cohesive subgroups, to identify groups of words that have relatively strong, frequent and direct links between them. These subgroups represent the themes of the reports in each subnetwork. Using the adjusted Rand index we compared partitions, identified with the Louvain method and VOS Clustering. Based on E–I indices, we found that islands identify cohesive subgroups in the selected subnetworks better than Louvain and VOS Clustering methods with the resolution parameter γ=1.
|
