Skip to main content
. 2021 Aug 5;2(3):e29570. doi: 10.2196/29570

Figure 1.

Figure 1

Illustration of the proposed Retweet-BERT. We first fine-tuned it on the retweet network (left) using a Siamese network structure, where the two BERT networks share weights. We then trained a denser layer on top to predict polarity (right).