Benchmarking of Transformer-Based Pre-Trained Models on Social Media Text Classification Datasets