\documentclass[twocolumn]{article}

\usepackage{algorithm2e}
\usepackage{showframe}

\begin{document}

\begin{algorithm}%[H]
{\small
\SetAlgoLined
\SetKwInOut{Input}{Input}
\SetKwInOut{Output}{Output}
\SetKwInOut{Assign}{Assign}
\SetKwInOut{Initialization}{Initialization}
\SetKwInOut{Iteration}{Iteration}
\SetKwInOut{Require}{Require}
\SetKwInOut{End}{End}
\SetKwInOut{Solution}{Solution}
%\Require{} ${\mathbf \alpha}, \beta, {\mathbf \gamma}_d$, Corpus{\small
\Input{${\mathbf \alpha}, \beta, {\mathbf \gamma}_d$, Corpus (labeled and unlabeled)}
%\Output{Ticket-topic matrix $\mat{U} \in {\mathbb R}_{\geq 0}^{n \times k}$, term-topic matrix $\mat{V} \in {\mathbb R}_{\geq 0}^{m \times k}$ and concept-topic matrix $\mat{W} \in {\mathbb R}_{\geq 0}^{h \times k}$}
\Initialization{ Initialize ${\bf \alpha} = (\alpha_1, \alpha_2, \ldots, \alpha_k)$\;
                             \hspace*{2.2cm}${\beta}$\;
                                                      \hspace*{2.2cm}$\gamma = (\gamma_1, \gamma_2, \ldots, \gamma_d)$\;
Initialize $D \times T$ matrix ${\mathbf \Theta}$, $D \times T \times S$ matrix ${\mathbf \Pi}$, $T \times S \times V$ matrix $\Phi$;}
\For{i = 1 to {\em max} Gibbs sampling iterations}
     { \For{all documents $d \in \{1,2, \ldots, D\}$}
        { \For {all words $w_t,\, t \in \{1,2, \ldots, N_d\}$}
                        {Exclude $w_t$ associated with topic $j$ and sentiment label $k$ from variables $N_{d,j}, N_{d,j,k}$, $N_{j,k,i}, N_j,k$ and $N_d$\;
                        Sample a new topic sentiment pair $\bar{z}$ and $\bar{l}$ using Eqn.~\ref{computedJointpWTS}\;
               Update variables $N_{d,j}, N_{d,j,k}$, $N_{j,k,i}, N_j,k$ and $N_d$ using the new topic label $\bar{z}$ and sentiment label $\bar{k}$\;
                                                                    }
                                        }
                                        \For{every 5 iterations}
                                         {Update hyperparameter $\bar{\alpha}$ with the maximum likelihood estimation according to Eqn~\ref{mAlpha} as discussed below}
                                        \eIf{number of iterations $\geq$ {\em max} Gibbs sampling iterations}
                                        {Update the matrices $\Theta, \Pi$ and $\Phi$ with new sampling results given by Eqns~\ref{Mtheta},~\ref{Mpi} and~\ref{Mvarpsi}  }{{\rm True}}
                                    }
    %\End{}
%\Return{\mat{V}, \mat{W}$\;}
}
 \caption{Gibbs sampling procedure for LJST}
\end{algorithm}

\end{document}