Temporal Classification and Visualization of Topics in a Twitter Search Interface

Searching within Twitter is a challenging task; the short and cryptic nature of tweets leads to search results sets that may include information on many different topics. While many topic modelling approaches exist to extract the salient topics from the tweets, what is missing is a method for temporally classifying the topics and showing these to a searcher to help them understand the makeup of the search results. In this work, we model the temporal distribution of the tweets that match each extracted topic, and then classify the topic as either emergent, stable, waning, or unclassified. This classification is visualized along side each topic, giving the searcher an easy way of identifying how the topics are changing over time. An example of this approach within an existing mobile Twitter search interface is provided.