MIT Press

Table 2

Summary of transfer operations defined in the compared models, where we assume the direction of transfer is from ℓ₁ to ℓ₂.

Model	Document level	Word level	Parameters of h
LDA	α	$β^{(ℓ_{2})}$	—

doclink	$δ \cdot N^{(ℓ_{1})} + α$	$β^{(ℓ_{2})}$	δ: indicator vector;

	$δ \cdot N^{(ℓ_{1})} + α$ ⁠,		$N^{(ℓ_{1})}$ ⁠: doc-by-topic matrix;

c-bilda	$δ \cdot N_{k}^{(ℓ_{1})} + χ^{(d)}$	$β^{(ℓ_{2})}$	supervision: comparable documents;
			δ: transfer distribution;

softlink	$δ \cdot N^{(ℓ_{1})} + α$	$β^{(ℓ_{2})}$	$N^{(ℓ_{1})}$ ⁠: doc-by-topic matrix;
			supervision: dictionary;
			δ: indicator vector;

voclink	α	$δ \cdot N^{(ℓ_{1})} + β^{(r, ℓ_{2})}$	$N^{(ℓ_{1})}$ ⁠: node-by-word matrix;
			supervision: dictionary;

Model	Document level	Word level	Parameters of h
LDA	α	$β^{(ℓ_{2})}$	—

doclink	$δ \cdot N^{(ℓ_{1})} + α$	$β^{(ℓ_{2})}$	δ: indicator vector;

	$δ \cdot N^{(ℓ_{1})} + α$ ⁠,		$N^{(ℓ_{1})}$ ⁠: doc-by-topic matrix;

c-bilda	$δ \cdot N_{k}^{(ℓ_{1})} + χ^{(d)}$	$β^{(ℓ_{2})}$	supervision: comparable documents;
			δ: transfer distribution;

softlink	$δ \cdot N^{(ℓ_{1})} + α$	$β^{(ℓ_{2})}$	$N^{(ℓ_{1})}$ ⁠: doc-by-topic matrix;
			supervision: dictionary;
			δ: indicator vector;

voclink	α	$δ \cdot N^{(ℓ_{1})} + β^{(r, ℓ_{2})}$	$N^{(ℓ_{1})}$ ⁠: node-by-word matrix;
			supervision: dictionary;