In this paper, we leverage the GPT-3.5 language model both using the Chat-GPT API interface and the GPT-3.5 API interface to generate realistic examples of anti-vaccination tweets in Dutch with the aim of augmenting an imbalanced multi-label vaccine hesitancy argumentation classification dataset. In line with previous research, we devise a prompt that, on the one hand, instructs the model to generate realistic examples based on the gold standard dataset and, on the other hand, to assign multiple pseudo-labels (or a single pseudo-label) to the generated instances. We then augment our gold standard data with the generated examples and evaluate the impact thereof in a cross-validation setting with several state-of-the-art Dutch large language models. This augmentation technique predominantly shows improvements in F1 for classifying underrepresented classes while increasing the overall recall, paired with a slight decrease in precision for more common classes. Furthermore, we examine how well the synthetic data generalises to human data in the classification task. To our knowledge, we are the first to utilise Chat-GPT and GPT-3.5 for augmenting a Dutch multi-label dataset classification task.

We present CoNTACT: a Dutch language model adapted to the domain of COVID-19 tweets. The model was developed by continuing the pre-training phase of RobBERT (Delobelle et al., 2020) by using 2.8M Dutch COVID-19 related tweets posted in 2021. In order to test the performance of the model and compare it to RobBERT, the two models were tested on two tasks: (1) binary vaccine hesitancy detection and (2) detection of arguments for vaccine hesitancy. For both tasks, not only Twitter but also Facebook data was used to show cross-genre performance. In our experiments, CoNTACT showed statistically significant gains over RobBERT in all experiments for task 1. For task 2, we observed substantial improvements in virtually all classes in all experiments. An error analysis indicated that the domain adaptation yielded better representations of domain-specific terminology, causing CoNTACT to make more accurate classification decisions. For task 2, we observed substantial improvements in virtually all classes in all experiments. An error analysis indicated that the domain adaptation yielded better representations of domain-specific terminology, causing CoNTACT to make more accurate classification decisions.