This paper describes the approach that we used to take part in the multi-label multi-class emotion classification as Track 3 of the WASSA 2023 Empathy, Emotion and Personality Shared Task at ACL 2023. The overall goal of this track is to build models that can predict 8 classes (7 emotions + neutral) based on short English essays written in response to news article that talked about events perceived as harmful to people. We used OpenAI generative pretrained transformers with full-scale APIs for the emotion prediction task by fine-tuning a GPT-3 model and doing prompt engineering for zero-shot / few-shot learning with ChatGPT and GPT-4 models based on multiple experiments on the dev set. The most efficient method was fine-tuning a GPT-3 model which allowed us to beat our baseline character-based XGBoost Classifier and rank 2nd among all other participants by achieving a macro F1 score of 0.65 and a micro F1 score of 0.7 on the final blind test set.

Annotating workplace bias in text is a noisy and subjective task. In encoding the inherently continuous nature of bias, aggregated binary classifications do not suffice. Best-worst scaling (BWS) offers a framework to obtain real-valued scores through a series of comparative evaluations, but it is often impractical to deploy to traditional annotation pipelines within industry. We present analyses of a small-scale bias dataset, jointly annotated with categorical annotations and BWS annotations. We show that there is a strong correlation between observed agreement and BWS score (Spearman’s r=0.72). We identify several shortcomings of BWS relative to traditional categorical annotation: (1) When compared to categorical annotation, we estimate BWS takes approximately 4.5x longer to complete; (2) BWS does not scale well to large annotation tasks with sparse target phenomena; (3) The high correlation between BWS and the traditional task shows that the benefits of BWS can be recovered from a simple categorically annotated, non-aggregated dataset.