*5.8. Synthetic Node Visualization*

9.25

(%)

Macro-F1

(%)

To analyze the synthetic nodes generated by **SORAG***F* and how they differ from the real nodes more intuitively, we projected the feature vectors of both the real and virtual nodes generated by **SORAG***F* for all datasets into two dimensions, as illustrated in Figures 8–10. For the dimensional-reduction method, we used t-SNE [56].

 52.60

**Figure 8.** A t-SNE visualization of synthetic and real node features of the BLOGCATALOG3 dataset obtained from **SORAG***F*. The node colors denote the labels.

**Figure 9.** A t-SNE visualization of synthetic and real node features of the FLICKR dataset obtained from **SORAG***F*. The node colors denote the labels.

**Figure 10.** A t-SNE visualization of synthetic and real node features of the YOUTUBE dataset obtained from **SORAG***F*. The node colors denote the labels.
