The paper discusses the prevalence of temporal data distribution shift in financial text and its impact on financial sentiment analysis. The authors conduct an empirical study using real-world financial social media data spanning three years. They find that fine-tuned models experience performance degradation when temporal distribution shifts occur. To address this issue, they propose a novel method that combines out-of-distribution detection with time series modelling. This approach is shown to enhance the model’s ability to adapt to shifting temporal conditions in volatile financial markets.
Publication date: 20 Oct 2023
Project Page: Not provided
Paper: https://arxiv.org/pdf/2310.12620
