Unpaired Sentiment-to-Sentiment Translation: A Cycled Reinforcement Learning Approach

• DOI: 10.48550/arxiv.1805.05181
• Publication Date: 2018-01-01
• Authors: Jingjing Xu; Xu Sun; Qi Zeng; Xuancheng Ren; Xiaodong Zhang; Houfeng Wang; Wenjie Li
• Keywords: Sentiment Analysis

The goal of sentiment-to-sentiment "translation" is to change the underlying sentiment a sentence while keeping its content. main challenge lack parallel data. To solve this problem, we propose cycled reinforcement learning method that enables training on unpaired data by collaboration between neutralization module and an emotionalization module. We evaluate our approach two review datasets, Yelp Amazon. Experimental results show significantly outperforms state-of-the-art systems. Especially, proposed substantially improves content preservation performance. BLEU score improved from 1.64 22.46 0.56 14.06 respectively.