Data Augmentation for Sentiment Analysis Using Sentence Compression-Based SeqGAN With Data Screening

Sentiment analysis refers to the process of automatically identifying emotions expressed by people. Its accuracy is highly dependent on amount training data. However, it takes time and cost for humans collect a large number Many research works used generative models generate data based small sentiment analysis. long texts inaccurate information that might be generated are two severe challenges. It difficult improve effectively. In this paper, we propose novel augmentation framework Sequence adversarial networks (SeqGAN) when dataset already has certain contains texts. Penalty-based SeqGAN high-quality diversified text Long short-term memory (LSTM) with attention mechanisms conduct sentence compression SeqGAN. A dictionary retain words compressed We also screening method obtain more accurate from The results usability, novelty, diversity show proposed can help learn classification four classifiers applied distinct datasets.