A Topic Modeling based Approach for Mining Online Social Media Data

Abstract

Dialogues authored by customers, or other social media users on social media platforms, provide focus into their perceptions, opinions, sentiments and concerns. Over the years, a huge extent of development in social media data has taken place. Examining this social media data has become a gigantic challenge. It is critical for businesses to grasp worthy knowledge from their customers, by gathering and surveying data produced by users on social media. Also if the organizations want to concentrate on any one specific aspect, they have to scan through the whole dataset, even though they needed to have access only to a few opinions/reviews of interest. So here we propose an approach such that the organization is able to deeply analyze, understand and gain knowledge from the data. We go about in the following way to mine online social media data. 1] Gathering the data 2] Topic Modelling 3] Classification and 4] Sentiment Analysis on the data. Here we present an approach that attempts to make the process of mining online data dynamic. Topic modelling here helps make in depth study about only one particular domain and get deeper insights regarding the same. We study two different topic modelling algorithms: Latent Dirichlet Allocation (LDA) and Non Negative Matrix Factorization (NMF). Also we attempt to make an improvement over the standard LDA algorithm for optimizing clusters, by integrating it with K medoids clustering algorithm.