Topic Modeling as a Tool to Gauge Political Sentiments from Twitter Feeds

Topic Modeling as a Tool to Gauge Political Sentiments from Twitter Feeds

Debabrata Sarddar, Raktim Kumar Dey, Rajesh Bose, Sandip Roy
DOI: 10.4018/978-1-6684-7472-3.ch031
(Individual Chapters)
No Current Special Offers


As ubiquitous as it is, the Internet has spawned a slew of products that have forever changed the way one thinks of society and politics. This article proposes a model to predict chances of a political party winning based on data collected from Twitter microblogging website, because it is the most popular microblogging platform in the world. Using unsupervised topic modeling and the NRC Emotion Lexicon, the authors demonstrate how it is possible to predict results by analyzing eight types of emotions expressed by users on Twitter. To prove the results based on empirical analysis, the authors examine the Twitter messages posted during 14th Gujarat Legislative Assembly election, 2017. Implementing two unsupervised clustering methods of K-means and Latent Dirichlet Allocation, this research shows how the proposed model is able to examine and summarize observations based on underlying semantic structures of messages posted on Twitter. These two well-known unsupervised clustering methods provide a firm base for the proposed model to enable streamlining of decision-making processes objectively.
Chapter Preview


The emergence of micro blogging sites has ushered in an immense wave of outpouring of public sentiment, views, and opinions. Peoples across the globe can, with a few keystrokes, make their views known to others of diverse social, religious, cultural and geographic backgrounds. No longer is the word of mouth or ideas confined to a localized section of society. The power and reach of the almost all-encompassing worldwide web have now made it possible for millions to view a microblog in seconds and at almost a press of a button irrespective of geographical distances involved.

Analysis of tweets related to politics can reveal trends. Evaluation of trends is imperative for ascertaining breaking news, generate friend recommendations, conduct data mining on opinions espoused by users, and record sentiments (Alvarez-Melis, 2016). Natural Language Processing (NLP) allows to classify opinions expressed through posts on Twitter and Facebook. NLP is comprised of opinion mining and text mining (Zhang & Liu, 2017). The impact of mining data from microblogging sites is significant. Insofar as analyzing the views on new musical releases, new product launches, motion picture releases, and similar other entertainment features, mining of tweets has been shown to be instrumental in helping formulate strategies accordingly (Mohey & Mohamed, 2016).

Complete Chapter List

Search this Book: