Geo-spatial-based Emotions: A Mechanism for Event Detection in Microblogs

Abstract

The use of emotions in microblogs to trace the occurrence of certain events and determine their locations is an open challenge for sentiment analysis. This study investigated the potential of detecting the geographical location of events based on existing linkages between the types of emotion embedded in tweets (degree of polarity) and the source location of those tweets. The extracted tweets were clustered using K-means algorithm and a predictive model was developed using Naïve Bayes algorithm. Then, a time series forecasting technique was applied using linear regression analysis. This method was used to predict the amount of emotions in association with the event of interest. Latitude and longitude were used to evaluate the results of the linear regression model on a real-time world map. Results showed that happy emotion tends to be a reliable source for detecting the geographical location of an event. This study revealed the feasibility of using the time series forecasting approach in investigating the degree of emotions in twitter messages.