TABLE 4.

Details of selected KGE algorithms: class of algorithm, triple scoring functions and their space and time complexities. Notation used: $m=|\mathcal{N}|$ , $n=|\mathcal{R}|$ , n _t = # of training triples, $\mathbf{h},\mathbf{r},\mathbf{t}\in {\mathbb{R}}^d$ , τ = |Ω| = # of convolution filters.

Algorithm	Class	Scoring function	Space (S) and time (T) complexity
TransE	Geometric	f r (h, t) = − |h + r − t|1/2	S: $\mathcal{O}(md+nd)$ T: $\mathcal{O}(n_{t}d)$
HolE	Matrix factorization	$[\mathbf{h}\star \mathbf{t}]={\sum }_{k=0}^{d-1}{[\mathbf{h}]}_k.{[\mathbf{t}]}_{(k+i)\mathrm{mod}d}$ f r (h, t) = r T (h ⋆ t)	S: $\mathcal{O}(md+nd)$ T: $\mathcal{O}(n_{t}d\log d)$
ConvKB	Deep learning	f r (h, t) = concat (g ([h, r, t]∗Ω)).W	S: $\mathcal{O}(md+nd+(\tau +3)d)$ T: $\mathcal{O}(n_t\tau d)$