Modeling crude oil volatility using economic sentiment analysis and opinion mining of investors via deep learning and machine learning models

Investor emotions can significantly impact the perceived value of financial assets, and sentiment analysis technology can accurately model daily responses and views of investors. In this study, we analyze investor remarks from one of the most popular web economic areas to investigate the effects of investor emotions on the price of crude oil in China. We construct investor emotions indices for China's crude oil futures based on specific economic factors and employ five machine learning technologies, including the AR approach, SVR approach, MLP method, neural networks with a recurrent units model, a recurrent unit with a gate approach, and LSTM model, to predict one-day-ahead prices. Our research offers insights into the effectiveness of different machine learning technologies and how specific economic factors contribute to sentiment analysis accuracy. We discuss the natural language processing techniques and tools used to extract sentiment from financial forum comments, along with the measures taken to ensure the accuracy and reliability of sentiment scoring. Our study highlights the advantages of the LSTM model, paired with the compound sentiment index, in the context of crude oil price forecasting and specifies the accuracy levels achieved by this model in comparison to others. We also discuss potential limitations and challenges in the long-term application of sentiment analysis for price forecasting and propose possible solutions. Our findings have practical implications for investors, financial analysts, and policymakers involved in China's crude oil market. We explore patterns or trends in investor emotions in our analysis and discuss their correlation with major economic or geopolitical events in China and globally. Our study contributes to a deeper understanding of investor emotions' impact on crude oil prices in China and provides valuable insights for various stakeholders in the dynamic and speculative energy oil industry. •We introduce method of calculation of investor Sentiment for crude oil prices.•We analyze investor remarks from web economic areas to investigate the effects of investor emotions on the price of crude oil.•Five well-known machine learning/deep learning algorithms used for analysis.•LSTM paired with the compound sentiment index performed exceptionally well.