Supervised Cross-Momentum Contrast: Aligning representations with prototypical examples to enhance financial sentiment analysis

Bo Peng, Emmanuele Chersoni, Yu-Yin Hsu, Le Qiu, Chu-Ren Huang

Abstract: Highlights•Cross-contrasting financial and prototypical representations enhances performance, even without complete comprehension of financial text.•The memory queue and momentum update mechanism enhance learning by allowing one anchor representation to be simultaneously contrasted with multiple distinct representations.•Supervised contrastive learning improves classification performance by aligning text representations based on sentiment categories.