Font Size: 

Spammers and spammers abound on social networks. Despite the fact that social media platforms have implemented a number of techniques to prevent spam from spreading, tight information review mechanisms have given rise to more sophisticated spammers. In this paper, we present a spam detection approach based on the self-attention Bi-LSTM neural network model in combination with ALBERT, which is word vector model. We use ALBERT to convert text from social networks into word vectors, which we then feed into the Bi-LSTM layer. The final feature vector is created after feature extraction and combining it with the self-attention layer's information focus. Finally, the result is classified by the SoftMax classifier.