采用生成对抗网络的金融文本情感分类方法

针对目前金融领域文本存在标注资源匮乏的问题，提出一种基于生成对抗网络的金融文本情感分类方法. 该方法以边缘堆叠降噪自编码器生成鲁棒性特征表示作为输入，在生成对抗过程中，通过向文本表示向量添加噪声向量再生成新样本，应用对抗学习思想优化文本特征表示. 在公开的跨领域情感评论Amazon数据集和金融领域数据集上进行实验，并与基准实验对比，结果表明，该方法在平均准确率上有显著提升.

Financial text sentiment classification based on generative adversarial network

There is a shortage of labeling resources in the texts of the financial field today.To address these issues, this paper presents a cross-domain text sentiment classification method based on Generative Adversarial Network. The method uses the Marginalized Denosing Autoencoders (mSDA) to generate a robust feature representation as input. In the process of generating adversarial, by adding to the text representation vector The noise vector is regenerated to generate a new sample, and the anti-learning idea is applied to optimize the text feature representation. Experiments were conducted on public cross-domain sentiment reviews on Amazon datasets, which showed a significant improvement in average accuracy compared to benchmark experiments.