Research On Soybean Price Forecasting Method Based On Network News

Soybean is the fourth largest food crop with high quality and high nutrition in China and even in the world,which has high economic and strategic value.The fluctuation of Soybean price affects the price of other grains,meat prices and the stability of agricultural markets.Forecasting the soybean price as accurately as possible can provide reference for the government to make corresponding decisions according to the soybean price prediction,and to a certain extent,it can prevent the loss of related interest groups caused by the abnormal fluctuation of soybean price.The soybean price in China is affected by the international market,the frequent occurrence of international events in recent years increases the difficulty of soybean price prediction.The research purpose of this paper is to establish a new soybean price prediction model on the basis of soybean news analysis,and use deep learning method to forecast soybean price.The main work includes:Firstly,for the text vectorization of network news,a soybean news text vectorization model based on Paragraph Vectors is proposed to analyze the news text content related to soybean.Paragraph Vector model is used to vectorize soy-related news text into a form that can be input into the prediction model.Secondly,a sentiment classification algorithm based on the Bi-Level Attention Model(BAM)is proposed to analyze the sentiment expressed in soybean news texts.BAM was used to divide news into positive and negative ones.Positive news was represented as 1 and negative news as-1,forming an sentiment value sequence of soybean news.BAM was better than the other six models in the classification results of the test,and the classification accuracy of the two data sets reached88.64%and 89.32%,respectively.Then,the long short-term memory network was used to forecast the soybean price multivariate dataset based on network news and the soybean price univariate dataset,so as to compare the experimental results.Four evaluation indexes are selected to evaluate the performance of the model.The four evaluation indicators are MAE,RMSE,MAPE and adjusted R~2.Finally,this paper takes the soybean price of Heilongjiang Province as an example to make an empirical study on the forecasting method of soybean price based on news.Firstly,we use web crawler technology to get soybean market price from August 13,2018 to September 13,2020 in Heilongjiang Province,as well as news related to soybean during the same period.Secondly,the text vectorization and sentiment analysis of soybean news were carried out to get the text vector sequence and sentiment value sequence about soybean.A soybean multivariate data set was established,including soybean price,soybean news text vector and soybean news sentiment.Finally,this multivariate data set is used as the input of the Long Short-Term Memory(LSTM)model to predict the soybean price.This paper used LSTM to forecast the multivariate data of soybean price based on news and univariate data of soybean price respectively,and compared the experimental results with other models.The final results show that the multivariate prediction of soybean price based on news is better than the univariate prediction of soybean price.In the multivariate forecast results of soybean price based on news,RMSE is 0.15,MAE is 0.08,MAPE is 3.03%,and adjusted R~2is 0.81.This shows that the forecast of soybean price based on news can indeed improve the accuracy of soybean price forecast.This paper discusses the method of news text analysis for long text,and it is the first time to apply the price prediction model based on network news to the research field of soybean price fluctuation.It is also the first time to put forward a new model of soybean price prediction by combining network news analysis with LSTM model.The empirical results show that the price forecasting model based on news analysis is feasible in soybean price.This provides a new technical support for soybean price forecast.At the same time,the model can also be applied to other agricultural products under similar conditions,providing some reference ideas for the price regulation of agricultural products market in China.