Fig. 2
From: From free text to clusters of content in health records: an unsupervised graph partitioning approach
Planar layouts using the ForceAtlas2 algorithm (Jacomy et al. 2014) of some of the similarity graphs generated from the dataset of 3229 records. Each node represents a record and is coloured according to its hand-coded, external category to aid visualisation of the structure. Note that the external categories are not used to produce our content-driven multi-resolution clustering in Fig. 3. a Layout for the full, weighted normalised similarity matrix \(\hat {S}\) without MST-kNN applied. b–e show the layouts of the graphs generated from the data with the MST-kNN algorithm with an increasing level of sparsity: k=17,13,5,1 respectively. The structure of the graph is sharpened for intermediate values of k, and we choose k=13 for our analysis here