Emoji Helps! A Multi-modal Siamese Architecture for Tweet User Verification

Abstract

In the current paper, we have proposed a new multi-modal authorship verification approach for social media texts. Authorship verification is a task of verifying whether an unknown text is written by a suspect or not. Use of social media like Facebook and Twitter is increasing day by day because of digitization. People have grown accustomed to regularly post or tweet about their everyday life, memorable incidences, random thoughts, opinions, and much more. Emojis are widely used in these tweets and posts. The writing style of a user can differ from others, since word choices, sentence structures, usage of punctuation symbols, and use of emoji can be different. We have applied a multi-modal Siamese-based framework for automatic extraction of features from the given texts and emojis. After the extraction of features, the extracted features are applied to a neural network–based architecture for binary classification. A multi-modal Twitter-based dataset is created for evaluating the performance of the proposed framework. We obtained an average accuracy of 61.56% with 78.08%, 61.50%, and 58.32% precision, recall, and f-measure values, respectively.