A Financial Text Sentiment Analysis Method Combining Self-Supervised Learning

To accurately analyze sentiment polarity in financial texts, such as financial news and company announcements, we propose a sentiment analysis method that integrates selfsupervised learning. This approach consists of two main components: a self-supervised learning module and a supervised learning module. The self-supervised module annotates semantic roles in financial texts and employs a zero vector to randomly replace these roles, implementing a masking operation that requires the model to predict the masked roles from a candidate space. The supervised module classifies sentiment using a dualchannel architecture that combines BiLSTM and an enhanced Text-CNN. The overall task loss is computed as the sum of the losses from both modules. Experimental results on a selfconstructed Chinese financial sentiment analysis dataset demonstrate that the proposed method outperforms baseline models, achieving state-of-the-art performance.