Context Attention Heterogeneous Network Embedding

Table 5

Average length of context node sequences when extracting different ratios of edges.

% training edges	15%	35%	55%	75%	95%
HEP-TH	0.8	2.2	5.7	6.9	7.2
Cora	2.6	4.3	7.9	15.2	17.7
Zhihu	2.3	4.1	6.6	9.7	10.3