Skip to main content
Figure 15 | EPJ Data Science

Figure 15

From: The shocklet transform: a decomposition method for the identification of local, mechanism-driven dynamics in sociotechnical time series

Figure 15

Topic network inferred from weighted shock indicator functions. At each point in time, words are ranked according to the value of their weighted shock indicator function and the top k words are taken and linked pairwise for an upper bound of \(\binom{k}{2}\) additional edges in the network; if the edge between words i and j already exists, the weight of the edge is incremented. The edge weight increment at time t is given by \(w_{ij,t} = \frac{R_{i,t} + R_{j,t}}{2}\), the average of the weighted shock indicator for words i and j, with the total edge weight thus given by \(w_{ij} = \sum_{t} w_{ij,t}\). After initial construction, the backbone of the network is extracted using the method of Serrano et al. [97]. The network is pruned further by retaining only those nodes i, j and edges \(e_{ij}\) for which \(w_{ij}\) is above the pth percentile of all edge weights in the backboned network. The network displayed here is constructed by setting \(k = 20\) and \(p = 50\), where size of the node indicates normalized page rank. Topics are associated with distinct communities, found using the modularity algorithm of Clauset et al. [98]

Back to article page