An Empirical Approach to Detection of Topic Bubbles in Tweets

Abstract

As Twitter usage increases worldwide, it becomes an important part of the big data ecosystem. People from across the globe tweeting and retweeting a large number of tweets instantaneously results in exponential growth of the information diffusion. This in turn can cause information bubbles. As data from Twitter and other microblogs are used in predictive analytics in many areas such as stock price prediction, varied interpretations of tweets would be necessary for adaptive diffusion analysis. This research is based on the fact that diffusion of information in Twitter has contributed to predictive analytics in other fields. Our research is concerned with the volume and velocity aspects of big data. We propose a new diffusion model for "topics" in twitter data. For the sake of simplicity, we assume (without providing formal reasoning) that all properties such as user interest, user familiarity, user curiosity, etc. affecting information diffusion are encoded in the tweet/retweet itself. We suggest a concept of potential for topics and build a time series for analysis. The definition of potential and time series allows to model different scenarios in information diffusion contributed by both volume and velocity of data (tweets). We also introduce the idea of bubble in information diffusion and propose a method to detect bubbles.