Profiling Celebrity Profession from Twitter Data
Kumar Gourav Das, Braja Gopal Patra, Sudip Kumar Naskar · 2021
Twitter is one of the most popular social media platforms which enables users from different walks of life to have two-way communications. User categorization in the Twitter platform categorizes a group of users based on their profiles, posts, and tweeting behaviors. The purpose of user categorization is to deliver relevant information to a specific class of users based on their interests. In this work, we perform user-level categorization of celebrities based on their professions. Different classification systems are developed to classify users into coarse-grained and fine-grained categories. The accuracy of the proposed models is evaluated on three datasets (Indian, non-Indian and combined dataset) using various machine learning and deep learning frameworks such as Naïve Bayes (NB), Support Vector Machine (SVM), Multilayer perceptron (MLP) and Convolutional Neural Network (CNN). Four different categories of features are used for this purpose - stylistic, hashtag, embeddings, and topic-based features. CNN turned out to be the best performing model for all three datasets. CNN produced accuracies of 83.66%, 87.51 %, and 84.02% on the Indian, non-Indian, and the combined datasets respectively. For the fine-grained classification task, maximum average accuracies of 80.34 % and 85.07 % are obtained using multilayer perceptron on the Indian and non-Indian datasets, respectively.