Research On Exchange Rate Data Forecast Based On LSTM And News Sentiment Analysis

With the deepening of economic globalization,the exchange rate,as an important financial instrument connecting economic and trade activities among countries,always affects the balance between domestic economic development and international financial environment,while the exchange rate itself is also affected by various factors such as political,economic,social and international situations.With the overlapping effects of global economic uncertainty and political risks,the exchange rate fluctuations have become unusually frequent and the degree of volatility has been increasing,making exchange rate forecasting particularly important in order to cope with financial risks and economic instability.However,the foreign exchange market is multivariate,nonlinear and highly dynamic,with the intricacies among various influencing factors of exchange rates makes the exchange rate forecasting extremely challenging.Traditional theoretical models of exchange rate determination assume that it is difficult to meet the requirements of changing economic factors in the market and effective forecasts is not possible,while traditional time series models of exchange rates rely heavily on linear relationships in the exchange rate series and are based on the assumption of smooth time series,making it difficult to accurately fit exchange rate fluctuations.In addition,the existing exchange rate forecasting models are mostly based on technical analysis of the exchange rate time series without incorporating external economic fundamentals into the model,making it difficult to respond to sharp fluctuations in the exchange rate in the short term.With the wide dissemination of Internet information and the rise of the field of data mining,it has become possible to mine emotional information related to exchange rate fluctuations from Internet news.Therefore,this paper proposes an exchange rate forecasting algorithm incorporating news sentiment analysis based on deep learning,which can effectively combine technical analysis and fundamental analysis to achieve the prediction of short-term exchange rate fluctuations.The main research work of this paper is as follows.(1)To address the situation that there are few Chinese datasets in the field of natural language processing,and even fewer related to the financial field,relevant news data are collected through crawlers.The news dataset used for news sentiment analysis in this paper is constructed through data cleaning and data annotation.(2)To address the problems of multiple meanings of words and contextual semantic dependency in Chinese texts,this paper proposes a BERT-LSTM based news sentiment analysis model which uses a BERT pre-training model to fine-tune the downstream sentiment analysis task on the news dataset to obtain deep semantic features of news texts,and then captures contextual association information through the LSTM model to classify news sentiment.The results show that the BERT-LSTM-based news sentiment analysis model has the best results in all metrics compared with benchmark models such as Word2 vec,Glove,Text CNN,and BERT.(3)In response to the existing exchange rate prediction models that only use exchange rate data as input variables cannot achieve the prediction of short-term exchange rate fluctuations,this paper takes the foreign exchange market as the research object,selects four mainstream currency pairs with RMB as the core in the market,illustrates the feasibility of integrating news sentiment solution by analyzing the distribution characteristics of exchange rate fluctuations and the correlation between news sentiment and exchange rate fluctuations,introduces the deep learning in The LSTM model in deep learning is introduced to extract features from exchange rate Kline data and news sentiment respectively.In addition,the sentiment information mined from the news is incorporated into the prediction model using only the exchange rate data as input variables,and three exchange rate prediction algorithm schemes incorporating sentiment analysis are proposed in this way.Based on experimental validation,the exchange rate prediction based on the LSTM model has the best effect when compared with several benchmark models,the solution of incorporating the daily news sentiment ratio as a quantitative sentiment indicator into the exchange rate prediction can effectively reduce the prediction error and improve the accuracy of predicting ups and downs in the prediction of four mainstream currency pairs,then an accurate prediction of short-term exchange rates is performed.The impact on the exchange rate forecasting results is also analyzed.