Curtin OCAI at WASSA 2023 Empathy, Emotion and Personality Shared Task: Demographic-Aware Prediction Using Multiple Transformers

Abstract

The WASSA 2023 shared task on predicting empathy, emotion and other personality traits consists of essays, conversations and articles in textual form and participants’ demographic information in numerical form. To address the tasks, our contributions include (1) converting numerical information into meaningful text information using appropriate templates, (2) summarising lengthy articles, and (3) augmenting training data by paraphrasing. To achieve these contributions, we leveraged two separate T5-based pre-trained transformers. We then fine-tuned pre-trained BERT, DistilBERT and ALBERT for predicting empathy and personality traits. We used the Optuna hyperparameter optimisation framework to fine-tune learning rates, batch sizes and weight initialisation. Our proposed system achieved its highest performance – a Pearson correlation coefficient of 0.750 – on the conversation-level empathy prediction task1. The system implementation is publicly available at https://github.com/hasan-rakibul/WASSA23-empathy-emotion.