. 2021 Aug 13;2(3):146–156. doi: 10.1016/j.jnlssr.2021.08.002

Table 2.

Semi-supervised learning process.

Algorithm:DYGIE++ with Semi-Supervised Learning
Input: Training set of CPHIE dataset as seed corpus $(D_{s})$ unlabelled raw corpus $(D_{u})$
Output: Trained DYGIE++ model $(M_{K})$
Train DYGIE++ model $M_{0}$ on $D_{s}$ Randomly split $D_{u}$ into $K$ subsets
for $i$ in $1 . . . K$ do ${D_{u}}^{(i - 1) } \leftarrow$ Predict using $M_{i - 1}$ ${D_{u}}^{(i - 1)} \leftarrow$ Relabel ${D_{u}}^{(i - 1) }$ s.t. if $p (y_{k}^{}) < θ$ then* $y_{k}^{*} \leftarrow 0$
end Train DYGIE++ model $M_{i}$ on $D_{s} \leftarrow D_{s} + D_{u}^{(i - 1)}$ end return $M_{K}$