| With the development of our country's economy and society and the intervention of government forces,our country's futures trading market has gradually stepped onto the right track,and the forecast of futures trends has also attracted more and more attention from people and scholars.Existing attention-based hybrid neural network models often fail to consider the impact of sudden financial events on futures prices when predicting futures trends.Based on this subject,a short-term futures trend prediction model based on sentiment analysis of financial news is proposed.The data set used in this subject is the 5-minute crude oil main contract transaction data from March 18,2018 to March 18,2021,as well as the text information of futures news and comments at the corresponding time.Used methods based on data processing rules for the crawled news and comment data to de-duplicate the source data obtained from the source webpage,remove links,convert emoticon tags into corresponding texts,etc.,and label some of the data with emotional tags.Taking into account the wide application of the BERT language model and attention mechanism,as well as the excellent performance of the Fin BERT model obtained by pre-training the financial field data in the sentiment analysis task,the BERT-based financial news sentiment analysis model was designed.At the same time,the financial news sentiment analysis model is integrated into the attention-based hybrid neural network to form a short-term futures trend prediction model based on financial news sentiment analysis.Finally,the performance comparison of the CNN model,Bi LSTM model,Bi LSTM+Attention model,CNN+Bi LSTM+Attention model and the Fin BERT+CNN+Bi LSTM+Attention model based on better parameters on the data set of this subject are compared respectively.The experimental results show that the result of the futures trend prediction model based on financial news sentiment analysis proposed by the subject is better than other models. |
PDF Full Text Request