Twitter sentiment analysis for foreign exchange market prediction

• Main Author: Oon, Xue Ting
• Other Authors: Sourav Saha Bhowmick
• Format: Final Year Project
• Language: English
• Published: 2019
• Subjects: DRNTU::Engineering::Computer science and engineering
• Online Access: http://hdl.handle.net/10356/76955
• Institution: Nanyang Technological University

The goal of the project is to predict the result of the GBP/USD currency pair; whether the closing price is lower or higher than opening price, based on tweets collected. The tweets were from Twitter accounts with great influence such as politicians like the 44th and 45th president of the United States of America; Barack Obama and Donald Trump, as well as news outlets such as CNN, BBCWorld. Unlike Stock market which had several studies on predicting stock price or trend using sentiment analysis [1, 2], it is rare to see predictive text mining applying in the context of Foreign Exchange (FX) market. Thus, the main purpose of this project was to provide a quantitative approach to analyse qualitative data like tweet contents and predict the result of the currency pair. The project was divided into five phases: 1. Data Acquisition Tweets and GBP/USD rate were collected using the Twitter and Quandl AP respectively. 2. Data Cleaning Cleaning of tweets context such as removing words, symbols that have no sentiment value and reducing remaining words to its root form were performed. This would help in reducing the text noise to a certain extent. 3. Data Exploration Before performing data modelling, it is important to understand the characteristics of the involved data. This would help in discovering any underlying relationship between the datasets, and thus lead to discovering new attributes. 4. Data Modelling Sentiment Analyser: Provides the sentiment value based on the tweet content, which will be used as input for the predictive model. Predictive Model: Predicts the result of the GBP/USD currency pair with supervised learning models such as Support Vector Machine. These models used the opening price of GBP/USD currency pair and the sentiment value as their inputs. Subsequently, evaluation of the above models would be done based on the predicted and actual outputs. Metrices like precision, recall and F1-Score were used to measure the accuracy of the models. 5. Data Visualization The findings of the project were displayed and presented through a dashboard. To conclude, the project could not predict the result of GBP/USD rates with high accuracy due to the lack of a strong relationship between tweets’ sentiment and FX rate. Lastly, the project would also discuss the other approaches used to analyse the tweets’ sentiment and improvement for future researches.