Sentiment Classification Based on RoBERTa and Data Augmentation
Xiaoyi Wang, Siyuan Xue, Jie Liu, Jing Zhang, Jincheng Wang, Jianshe Zhou · 2023
Sentiment analysis is one of the hot research topics in natural language processing, aiming to analyze the subjective sentiment expressed in text through inductive reasoning. Common deep learning-based sentiment classification models often require the integration of word embedding techniques. However, conventional approaches to generating word embeddings struggle to effectively capture the bidirectional semantic features of text. RoBERTa, with its unique attention mechanism and masked technique, enables the capture of more comprehensive features in the text. In this paper, we introduce the RoBERTa model to the task of Chinese comments sentiment classification and address the limited data issue by employing data augmentation methods to expand the dataset. Through a series of experiments conducted on three different datasets, we demonstrate the effectiveness of the sentiment classification model that combines RoBERTa with data augmentation methods.