Construction of Effective News Corpus in Commodity Futures Domain

Lingling Mu, Le Kang, Yintao Jia, Ning Zhao, Wenxin Li, Xiaoyang Feng, Hongying Zan

Abstract: Effective news in the domain of commodity futures contains information about market analysis and operation suggestions, which has a significant impact on futures prices. It is of great significance for the supervision and risk prediction of the futures market to identify the effective news and analyze its correlation with the futures market. This paper collected cotton futures news from January 2020 to March 2021 from major Chinese websites, and manually annotated 5,025 news passages to construct effective news corpus FENC (Future Effective News Corpus), which includes 2,828 effective news. We used FENC to train effective news classification models based on pre-trained models. This paper also ensemble SOTA classification models to automatically identify the effective news from January 2020 to March 2021, and automatically constructed the extended effective news corpus FENC-E which includes 34,272 effective news and 30,211 non-effective news. FENC and FENC-E can be used for the effective news identification in the futures domain.