Does pre-processing affect the correlation indicator between Twitter message volume and stock market trading volume?

Abstract

Motivation: More and more authors empirically verify the relationship between the volume of tweets and the stock market indicators. The patterns explored from Twitter most often take the form of time series that represent user's activity on different level of gran-ularity (moods, emotions, relevant topic or query-related messages). Sentiment analysis is a technique used to transform text data into information on the mood and related be-havioral categories. Supervised machine learning is the most commonly used approach to sentiment analysis. Thus, the results of an empirical analysis of the relationship between social media and stock depend on the quality of results of classification task. The quality of the features used to learn the classifier plays a key role. The feature space is modified using various data pre-processing scenarios that aim to increase accuracy of classification. The impact of pre-processing data on the quality of classification is often discussed in studies. Very few authors discuss the impact of pre-processing on the correlation indicator between Twitter and stock market. Aim: Analysis of the impact of tweets pre-processing on the Pearson correlation indicator between the mood of Twitter users and stock market trading volume. Results: The correlation between the volume of stock market trading and the volume of tweets has been empirically confirmed. The effect of pre-processing on the correlation index was noted for the variables 'all_tweets' and 'negative_tweets'. This is because the training set has a significant amount of tweets with negation. However, the results ORIGINAL ARTICLE