TY - JOUR
T1 - Analysis of timeliness of oil price news information based on SVM
AU - Zhao, Lu Tao
AU - Zeng, Guan Rong
N1 - Publisher Copyright:
© 2019 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/) Peer-review under responsibility of the scientific committee of ICAE2018 - The 10th International Conference on Applied Energy.
PY - 2019
Y1 - 2019
N2 - International oil price forecasting has always been an important and difficult issue. With the renovation of big data technology, it is a new idea to correct and improve the oil price forecasting model by collecting and extracting network news information. However, due to timeliness and other complex factors of the network news, the ability of online news to predict oil prices is unstable. In this paper, a novel method based on SVM (support vector machines) is proposed to explore the timeliness of news towards oil price. The timeliness of news can be explained by the accuracy of SVM. A multi-scale trend discovery method is proposed to more flexibly extract oil price trends at different scales and in different directions. The news texts are marked in 82 different ways. With the help of SVM, the text information is directed and extracted, and the ability of the news to describe fluctuations in oil price trends is reflected by the quality indicators of classification. The empirical results show that the new multi-scale trend discovery method has high reliability. The ability of news to describe the short-term oil price trend only lasts for about one week, and the relatively long-term trend can be maintained for two or even three weeks. News is more able to express information on relatively long-term trends. In addition, information described in the news is more likely to have an impact on the future rather than on the volatility of past oil prices. This study of the interactive relationship between news texts and oil price trends is a powerful support for the application of news texts to forecast oil prices. At the same time, it is also a powerful complement to the study of discrete correlations and interactions.
AB - International oil price forecasting has always been an important and difficult issue. With the renovation of big data technology, it is a new idea to correct and improve the oil price forecasting model by collecting and extracting network news information. However, due to timeliness and other complex factors of the network news, the ability of online news to predict oil prices is unstable. In this paper, a novel method based on SVM (support vector machines) is proposed to explore the timeliness of news towards oil price. The timeliness of news can be explained by the accuracy of SVM. A multi-scale trend discovery method is proposed to more flexibly extract oil price trends at different scales and in different directions. The news texts are marked in 82 different ways. With the help of SVM, the text information is directed and extracted, and the ability of the news to describe fluctuations in oil price trends is reflected by the quality indicators of classification. The empirical results show that the new multi-scale trend discovery method has high reliability. The ability of news to describe the short-term oil price trend only lasts for about one week, and the relatively long-term trend can be maintained for two or even three weeks. News is more able to express information on relatively long-term trends. In addition, information described in the news is more likely to have an impact on the future rather than on the volatility of past oil prices. This study of the interactive relationship between news texts and oil price trends is a powerful support for the application of news texts to forecast oil prices. At the same time, it is also a powerful complement to the study of discrete correlations and interactions.
KW - International Oil Price
KW - Text Mining
KW - Timeliness
KW - Trend
UR - http://www.scopus.com/inward/record.url?scp=85063915824&partnerID=8YFLogxK
U2 - 10.1016/j.egypro.2019.01.821
DO - 10.1016/j.egypro.2019.01.821
M3 - Conference article
AN - SCOPUS:85063915824
SN - 1876-6102
VL - 158
SP - 4123
EP - 4128
JO - Energy Procedia
JF - Energy Procedia
T2 - 10th International Conference on Applied Energy, ICAE 2018
Y2 - 22 August 2018 through 25 August 2018
ER -