MINT: A Tool to Explore Themes in High Velocity Social Media Data

EasyChair Preprint 669

Abstract

An important aspect of social media analytics is understanding themes, i.e., popular discussion topics, as they evolve in time. Techniques like topic models, which assume a generative model of theme distribution over documents, often suffer from three problems -- they are expensive to compute in real-time, they usually limit the number of topics, and they do not consider the continuous change in the topic proportions, emergence and noise that occurs in social media. To address these problems, we have developed MINT, a framework that uses a non-uniform stochastic model of term occurrence to identify themes as they evolve in real-time, and periodically stores the identified themes in a semi-structured database. MINT therefore supports both theme detection and evolution queries on social media. Themes encode useful information like their rate of growth, changing user communities around them, links to related real-world events, their representative topic-handles, etc., that can be queried upon. We experimentally show that our extracted themes compare well with more traditional methods but performs well in real-time.

Keyphrases: High velocity data, high velocity social media data, social media monitoring, theme component, theme extraction algorithm, theme set, trend analysis

@booklet{EasyChair:669,
  author    = {Nishant Agarwal and Subhasis Dasgupta and Amarnath Gupta},
  title     = {MINT: A Tool to Explore Themes in High Velocity Social Media Data},
  doi       = {10.29007/mmh3},
  howpublished = {EasyChair Preprint 669},
  year      = {EasyChair, 2018}}