Empirical Study of Tweets Topic Classification Using Transformer-Based Language Models

Abstract

Social media opens up a great opportunity for policymakers to analyze and understand a large volume of online content for decision-making purposes. People’s opinions and experiences on social media platforms such as Twitter are extremely significant because of its volume, variety, and veracity. However, processing and retrieving useful information from natural language content is very challenging because of its ambiguity and complexity. Recent advances in Natural Language Understanding (NLU)-based techniques more specifically Transformer-based architecture solve sequence-to-sequence modeling tasks while handling long-range dependencies efficiently, and models based on transformers setting new benchmarks in performance across a wide variety of NLU-based tasks. In this paper, we applied transformer-based sequence modeling on short texts’ topic classification from tourist/user-posted tweets. Multiple BERT-like state-of-the-art sequence modeling approaches on topic/target classification tasks are investigated on the Great Barrier Reef tweet dataset and obtained findings can be valuable for researchers working on classification with large data sets and a large number of target classes.