Utilising Sentiment Analysis To Predict Market Movements In Currency Exchange Rates

One of the most important parts of the international economy is the foreign currency market, which shows how different countries' economies are doing. Due to the Internet's growing popularity, more individuals are opting to articulate their thoughts and viewpoints online. Hence, we anticipate that the fluctuations in the currency exchange market will be mirrored in the textual remarks shared on the Internet. This article presents a hybrid model that integrates sentiment analysis to predict future trends in the currency market. The sentiment analysis module integrates Word2vec and Long Short-Term Memory to enhance the model's performance and extract "emotion" variables from the text input. The algorithm utilizes sentiment analysis to identify the emotions expressed in "exchange rate" posts on "Weibo" as either "positive," "ordinary," or "negative." It then assigns an emotional weight to each day based on these categorizations. This enables the model to ascertain the emotional significance of each day. The model generates a word vector once the data-cleaning process is finished. Next, the historical exchange rate data spanning three to four years was randomly combined with the emotional weights from roughly three months to undertake mixed training, which included CNN and LSTM models. The dataset that is being used for this analysis includes information regarding the United States dollar, the Japanese yen, the euro (EUR), and the Australian dollar in relation to the RMB exchange rate, with samples spanning three to four years. As measured by metrics such as R2_Score, Mean Absolute Percentage Error (MAPE), Mean Squared Error (MSE), and Mean Absolute Error (MAE), the experimental results show that the method presented in this study is better than other traditional models. Our technique predicts the Australian dollar/RMB exchange rate trend better than competing models by 38.41% in MAE, 34.42% in MSE, 3.08% in MAPE, and R2_Score. This shows our method's average difference from the models. The average score of 6.21%, which is higher than the comparative models, indicates a significant level of resilience. After extracting the recommended method from the "emotional analysis" module, we compared our results to see how well our emotional weights for future exchange rate forecasts worked. The findings demonstrated that the module integrating "emotional analysis" exhibited significantly superior performance compared to the module lacking this component.