Expression of Feelings in Twitter: A Decision Tree Approach

Abstract

Studies that examined the relationship between expressing feelings, such as feeling ‘bored’, ‘excited’, ‘lonely’, loved’, ‘sad’ and ‘happy’ and Twitter users’ network size (i.e. the number of friends and the number of followers) did not take into account the influence of other factors, such as the number of tweets, the number of lists and the number of favourites, because prior research did not provide clues as to why these should be considered. The data mining approach is not biased by prior knowledge. In this study a data mining technique, specifically a decision tree, was applied to look at the interaction between the expression of feelings and all Twitter users’ attributes considered likely to be useful in the discovery of interesting rules. The decision tree technique was applied on a large dataset of tweets containing the phrases, in double quotations marks, “I am bored”, “I am excited”, “I feel lonely”, “I feel loved”, “I feel sad” and “I feel happy”. Only when these phrases were tweeted twice or more at different times that they were retrieved from Twitter using the Digital Methods Initiative Twitter Capture and Analysis Toolset (DMI-TCAT). The decision tree technique generated a number of interesting rules that provided clues about previously unknown relationships between the expression of feelings and a number of Twitter users’ attributes. This study demonstrates that data mining is valuable for shedding light on previously unconsidered factors that can influence the expression of feelings; thereby advancing the research in this area.