Predicting NASDAQ Stock Prices Through Sentiment Analysis of U.S. President’s Social Media Posts

This project investigates the influence of U.S. President Donald Trump's social media posts on the NASDAQ Composite Index by leveraging sentiment analysis and deep learning-based time series prediction models. The code in this repository supports the experiments described in the draft paper, which aims to empirically validate how political sentiment impacts financial markets.

Using sentiment analysis tools like VADER and RoBERTa, the project extracts positive, neutral, and negative sentiment from the President's posts. The extracted sentiment data, along with historical NASDAQ index values, are used as inputs to Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) models to predict stock market trends.

This repository includes scripts for data preprocessing, model training, performance evaluation, and results visualization, providing a reproducible framework for the experiments presented in the draft paper.

Disclaimer

This repository contains code related to a draft paper that is currently under review. The details and results are subject to change. If you have any questions, please feel free to contact us.

Table of Contents

• Introduction
• Dataset
• Code Overview

Introduction

This repository accompanies a draft paper titled:
"Predicting NASDAQ Stock Prices Through Sentiment Analysis of U.S. President’s Social Media Posts"
(Currently under review, preprint not yet available)

The repository provides code for reproducing the core experiments described in the draft paper. Once the paper is finalized and published, the README will be updated to include a formal citation.

Dataset

The dataset used in this project is located in the data directory.

Data File	Description
text_and_sentiment.csv	Dataset with sentiment scores calculated using VADER and RoBERTa based on the text column.
sentiment_and_nasdaq.csv	Dataset created by performing an inner join with NASDAQ data based on the date column.

Code Overview

Below is a breakdown of the code files and their intended purposes, along with their relationship to the draft paper:

Code File	Paper Section	Description
01_TrumpCrawling.py	Section 4.1.2: Data Collection	Crawls Truth Social data using Beautiful Soup 4.
02_nasdaq.ipynb	Section 4.1.2: Data Collection	Collects NASDAQ data using the Finance Data Reader library.
03_preprocess.ipynb	Section 4.1.3: Data Preprocessing	Performs data preprocessing, such as converting dates to EST, removing URLs, and handling missing values.
04_EDA.ipynb	Section 4.1.3: Data Preprocessing	Conducts basic exploratory data analysis (EDA) on preprocessed data.
05_VADER and RoBERTa sentiment analysis.ipynb	Section 4.2.2: Utilizing Sentiment Analysis Results	Conducts sentiment scoring using VADER and RoBERTa.
06_data join.ipynb	Section 4.2.3: Time Series Prediction Model	Merges Trump data with NASDAQ data.
07_LSTM.ipynb	Section 5.1: LSTM	Trains and evaluates the performance of the LSTM model.
08_GRU.ipynb	Section 5.2: GRU	Trains and evaluates the performance of the GRU model.