Research On Crude Oil Price Prediction Model With Text Characteristics

As the primary energy with the largest proportion in the global energy,crude oil is an indispensable strategic resource for national survival and development,it plays an inestimable role in ensuring national economy and promoting social development.Therefore,the fluctuation of crude oil has a great impact on global economic development.Based on the importance of crude oil,the trend prediction of crude oil price is particularly critical for decision makers.With the diversified development of the media situation,the online market has produced more news events,among which the news about finance,foreign exchange and stock market is updated day by day.The information displayed in the news headlines will affect the market behavior of investors and show the fluctuation of the market through certain information.Therefore,combined with market data and news data,it is of certain significance to study the trend of crude oil price.This paper studies the prediction of crude oil price based on different forms of news data to compare the results of prediction using financial data.The research contents are as follows:1.Describing the method and source of obtaining data,and cleaning the format and content of data.Using outlier detection,missing value filling and data standardization to process financial data.Removing stop words and word segmentation of network news headline data.Finally,the relevant statistical analysis of the research data is carried out.2.The text data is processed from different levels of natural language processing,and then the crude oil price is predicted.Specifically,(1)in the text representation method,text embedding is selected to train the obtained text data and construct the word embedding model;(2)The text vector corresponding to the news headline is obtained through the model,and a new data set is obtained by combining the text vector with financial data;(3)Support vector regression,random forest,kernel ridge regression and lightgbm were used for prediction and analysis;(4)At the level of text analysis method,emotion analysis is selected to analyze the text emotion of news headline data,and its text emotion score is obtained;(5)The investor sentiment index is applied to the text sentiment score to calculate the investor sentiment value showing the psychological expectation of investors,and the sentiment score and investor sentiment value are combined with economic characteristics to predict.Comparing the model with text features with the benchmark model,it is found that the prediction accuracy has been improved to a certain extent.3.The news headlines are subject to subject classification,the sentiment score under each topic classification is calculated,and the sentiment score vector is obtained,which increases the dimension of the sentiment score,so as to obtain a better prediction effect.Specifically,(1)The traditional LDA topic model and short text BTM model are used to classify the news title data,determine the topic distribution of the news title,and segment the text emotion under the topic;(2)The text emotion under the topic classification is taken as the text feature,and the obtained financial market data is taken as the economic feature,which is input into four prediction models for prediction;(3)Through comparison,it is found that after the short text BTM model is applied to text emotion and features,its prediction accuracy is not only higher than that of the benchmark model,but also greatly improved compared with the prediction results of the prediction models based on text embedding,text coarse granularity emotion and LDA model.The above research shows that the prediction model added with the news headline text data of crude oil and its downstream products has better prediction effect on the prediction of crude oil price.Through comparative analysis,it is found that the short text BTM model combined with text emotion is the best for the prediction of crude oil price.