Algorithm 1: Online Algorithm for SimNest
	Input: Data matrix X = X1 ∪ X2, Twitter data stream C, contact network G.
	Output: the population’s predicted health states Ƶ.
1	Set the learning rate η = 0.5. Initialize weight matrix W as matrix of random values between -1 and 1;
2	repeat
3	Update unlabeled data set X2 by Twitter data stream;
4	repeat
5	Randomly select a labeled sample (Xu,t, Yu,t);
6	$W\leftarrow W-\eta \cdot \frac{\partial {ℒ}_1(X_{u,t},Y_{u,t},W)}{\partial W}$;
7	Randomly select an unlabeled sample Xu;
8	$W\leftarrow W-\eta \cdot \frac{\partial {ℒ}_3(X_u,p_I,W)}{\partial W}$;
9	Randomly select an unlabeled sample Xυ;
10	for i ← 1 to T do
11	$W\leftarrow W-\eta \cdot \frac{\partial {ℒ}_4(X_{\upsilon ,i},X_{\upsilon ,i+1},W)}{\partial W}$
12	end
13	Randomly select a user u from a location l ∈ L;
14	$W\leftarrow W-\eta \cdot \frac{\partial {ℒ}_2(X_{u,t},G,p_E,p_I,W)}{\partial W}$;
15	${\mu }_I\leftarrow \frac{1}{|U_2|}\sum _u^{U_2}\sum _t^T{f}_W(X_{u,t})$;
16	${\sigma }_I\leftarrow {({\sum }_u^{U_2}{\sum }_t^T{f}_W(X_{u,t}-{\mu }_I)/\left|U_2\right|)}^{1/2}$;
17	until converge;
18	$p_E,Ƶ\leftarrow \min \sum _t^T\sum _l^{ℒ}{\Vert {\mathrm{\lambda }}_1\sum _{\upsilon }^{V_l}{Q}_{\upsilon ,t}(G,p_E,p_I)-\sum _u^{U_{2,l}}{f}_W(X_{u,t})\Vert }^2$;
19	${\mathrm{\lambda }}_1\leftarrow \sum _{l,t}^{ℒ,T}\sum _u^{U_{2,l}}{f}_W(X_u,t)/\sum _{l,t}^{ℒ,T}\sum _{\upsilon }^{V_s}{Q}_{\upsilon ,t}(G,p_E,p_I)$
20	until the end of data stream;