Notation	Description
$N=\left\{n_1,n_2,\dots ,n_{\left|N\right|}\right\}$	Set of news items, $\left|N\right|$ is the size of the news dataset
$y_i\in \left\{0,1\right\}$	Label $y_i=1$ is fake news; $y_i=0$ is real news
$U=\left\{u_1,u_2,\dots ,u_{\left|U\right|}\right\}$	Set of users; $\left|U\right|$ is the number of users
$\left(u_0,s{c}_0,t_0\right)$	A tuple, user u and social context sc during timestamp t
$C\left(n_i\right)$	Content of news
$SC\left(n_i\right)=\left(\left(u_0,s{c}_0,t_0\right),\dots ,\right)$	Sequence of a user’ social contexts on a news item, |sc| is the size of SC
$\widehat{y}\left(n_i\right)\in \left\{0,1\right\}$	Predicted label for news item $n_i$
$\widehat{y}(n_i)=M\left(n_i,SC\left(n_i\right)\right)$	Model M predicts a label for news item based on its news features and social contexts
$X=\left\{x_1,x_2,\dots ,x_k\right\}$	Sequence of k input vector representations, k is the length
$X^{\prime }=\left\{x_1^{\prime },x_2^{\prime },\dots ,x_k^{\prime }\right\}$	Sequence of embedding vectors from X
$f:X_{1:k}^{\prime }\to Y_{1:l}$	Mapping f from input sequence of k vectors to a sequence of l target vectors
$x^{″}$, $X^{″}$	Output vector representation from input $x^{\prime },$ and sequence of output vectors of $x^{\prime }$
${\kappa }_i$; $v_i$; $q_i$;$\mathbb{K}$	Key vector; value vector; query vector; set of key vectors
$W_v$, $W_k$,$W_q$, SoftMax	Trainable weight vectors of $\kappa$; $v$; $q$, activation function
${\overline{X}}_{1:k};Y_{1:l}$	Contextualized input sequence to decoder; the target vector sequence
$f_{{\theta }_{\mathit{enc}}};p_{{\theta }_{\mathit{dec}}}$, ${\mathcal{L}}_{1:k}={\ell }_1,\dots ,{\ell }_k$	Encoder function, decoder function; logit vector
$y^{\prime }$; $y^{″}$	Vector representation of $y^{\prime }$, and $y^{″}$
< S > ; [$\mathcal{S}$];$h_{\left[\mathcal{S}\right]}$	Token in decoder; last state of the token; hien ste
$p\in {[0,1]}^2$	Probability distribution over classes [0,1]
$\mathcal{W}$; $b$; h	Project matrix; bias term; cross-entropy function
$X;X^{\prime }$; $X^{″}$;$\overline{X}$	Input sequence to encoder; sequence generated from $X$; sequence generated from $X^{\prime }$; output encoding sequence from $X^{″}$
$Y$; $Y^{\prime }$; $Y^{″}$; $Y^{″}$;$\overline{Y}$	Target sequence in decoder; sequence generated from $Y$; $Y^{\prime }$; $Y^{″}$; and $Y^{″\prime }$, respectively
$y_j$; ${\widehat{y}}_j$	Ground truth label; model prediction