| 講演抄録/キーワード |
| 講演名 |
2020-12-17 11:15
A Novel Data Augmentation Framework Based on SeqGAN for Sentiment Analysis ○Jiawei Luo・Mondher Bouazizi・Tomoaki Ohtsuki(Keio Univ.) PRMU2020-43 |
| 抄録 |
(和) |
(まだ登録されていません) |
| (英) |
Sentiment analysis is an important field in Natural Language Processing (NLP). It can analyze people's sentiment through their articles. On a related topic, machine learning has achieved high accuracy in sentiment analysis. However, it requires a large amount of high-quality training data that are hard to be collected. In this work, a novel data augmentation framework based on sequence generative adversarial networks (SeqGAN) is proposed to improve the sentiment analysis accuracy. In our framework, we conduct sentence compression and use a sentiment dictionary to retain the sentiment words for compressed data. The compressed data are used to train SeqGAN. We use the trained SeqGAN to generate artificial data for sentiment analysis. A classifier is used to discard generated data that may contain incorrect sentiment information. The results show that the proposed data augmentation framework helps SeqGAN generate high quality and novel text data. The data generated by the proposed framework improve the accuracy of the sentiment analysis classifier on some of the benchmark sentiment analysis dataset available. |
| キーワード |
(和) |
/ / / / / / / |
| (英) |
data augmentation / sentiment analysis / machine learning / sentence compression / SeqGAN / / / |
| 文献情報 |
信学技報, vol. 120, no. 300, PRMU2020-43, pp. 30-35, 2020年12月. |
| 資料番号 |
PRMU2020-43 |
| 発行日 |
2020-12-10 (PRMU) |
| ISSN |
Online edition: ISSN 2432-6380 |
著作権に ついて |
技術研究報告に掲載された論文の著作権は電子情報通信学会に帰属します.(許諾番号:10GA0019/12GB0052/13GB0056/17GB0034/18GB0034) |
| PDFダウンロード |
PRMU2020-43 |