HMBI: A New Hybrid Deep Model Based on Behavior Information for Fake News Detection

Algorithm 1

Training procedure of HBMI.

Input: news content and social behavior information , and labels {}.

Output: target model

1 for number of epoch do

2 through BERT pretrained model;

3 through BERT pretrained model;

4 compute according to Equation (2): ;

5 compute according to Equation (3): ;

6 linear projection;

7 for each encoder do

8 compute according to Equations (4) and (5):

9 ;

10 ;

11 end

12 compute according to Equation (6): ;

13 integrate and into : ;

14 compute according to Equation (7): ;

15 end

16 if the training converges then

17 ;

18 end

19 return ;