基于注意力机制的社交垃圾文本检测方法

在社交网络中，大量的垃圾文本严重威胁用户的信息安全与社交网站的信用体系。针对噪声性与稀疏性问题，提出一种基于注意力机制的卷积神经网络检测方法。在经典卷积神经网络的基础上，该方法增加了过滤层，并在过滤层设计基于朴素贝叶斯权重技术的注意力机制，解决了噪声性问题。并且，它改变了池化层原有的策略，采用基于注意力机制的池化策略，缓解了稀疏性问题。结果表明，相对于其他检测方法，所提方法的检测准确率在4个数据集上分别提高了1.32%、2.15%、0.07%、1.63%。

Attention-based approach of detecting spam in social networks

In social networks,a large amount of spam has seriously threaten users’information security and the credit system of social websites.Aiming at the noise and sparsity problems,an attention-based CNN method was proposed to detect spam.On the basis of classical CNN,this method added a filter layer in which an attention mechanism based on Naive Bayesian weighting technology was designed to solve the noise issue.What’s more,instead of the original pooling strategy,it adapted an attention-based pooling policy to alleviate the sparsity problem.Compared with other methods,the results show that the accuracy has increased by 1.32%,2.15%,0.07%,1.63%on four different data sets.