Exploring Customer comments using Latent Dirichlet Allocation

Abstract

With the advent of social media, a large number of customers share their experience on social media about the products & services that they consume. Detecting issues reported by customers is significant for providing solutions to their problems. As a result, there is an emerging sub-field of social media analytics to identify and comprehend customer feedback and understand topics hidden in it. Topic modeling algorithms, a branch of machine learning, extract relevant insights from customers' posts by identifying hidden words and patterns. This pilot study has been conducted to ascertain the most frequently discussed issues by users on social media microblogging site Twitter. We used Tf-idf to re-allocate the weights to feature words and Latent Dirichlet Allocation (LDA) for topic modeling. Additionally, we carried out a comparative study of LDA against GSDMM topic model.