Training a Neural Network to Perform Sentiment Analysis of Stocks-Related Tweets

Sentiment Analysis is a method of machine learning that creates opinion values based typically on text or other human-created documents. Any form of text written by a person or group can be evaluated with the values, consisting of positive, negative, and neutral. The impact that people's moods and opinions can have on stock prices can lead to accurate predictions of stock market movement. To find how this may affect the performance of a stock, we performed a sentiment analysis of stock-related Tweets to evaluate stakeholder's thoughts on the market and directly compare them. Information gathered from Yahoo Finance (a marketing site) is run alongside Tweets gathered from Twitter's API. The sentiment values of each Tweet are calculated through the use of a Long-Short Term Memory (LSTM) tool created to predict a relation between the selling prices of stocks and the company's sentiment. An F-1 score is calculated based on predictions and actual values which resulted in a 70% accuracy of predictions. This indicates that much more of the Tweets are neutral than previously realized. Although the F-1 score reads as such, the final graph and results of this research prove that there is in fact a correlation between stock sentiment values and daily closing prices, indicating the effects that opinions on Twitter, although small, affect stock prices.