Sentiment Analysis of Financial Media Commentary Text Based on FinBERT-BiGRU-CNN
Yiling Yan, Qiuli Qin · 2024
This study addresses the current gap in aspect-based sentiment analysis within financial media coverage sentiment research, as well as the incomplete feature extraction of financial commentary texts in existing models, and the absence of labeled datasets of domestic financial market commentary text for model training. To tackle these issues, we propose a FinBERT-BiGRU-CNN model that incorporates an attention-pooling strategy to capture aspectual entities within the middle layer of Fin-BERT, thereby enhancing the semantic, global, and local feature extraction of the text. Our proposed model achieves a notable F1 score of 0.8734 for sentiment classification. However, the comparative experiments are hindered by the scarcity of authoritative datasets suitable for Aspect-Based Sentiment Analysis (ABSA) tasks in the financial domain. In conclusion, the experimental results underscore the effectiveness of the FinBERT-BiGRU-CNN model for aspect-level sentiment analysis tasks within the financial domain.