Incorporating Multiple Knowledge Sources for Targeted Aspect-based Financial Sentiment Analysis

Combining symbolic and subsymbolic methods has become a promising strategy as research tasks in AI grow increasingly complicated and require a higher levels of understanding. Targeted Aspect-based Financial Sentiment Analysis (TABFSA) is one of such complicated tasks, as it involves information extraction, specification, and domain adaptation. External knowledge has been proven useful for general-purpose sentiment analysis, but not yet for the finance domain. Current state-of-the-art Financial Sentiment Analysis (FSA) models, however, have overlooked the importance of external knowledge. To fill this gap, we propose using attentive CNN and LSTM to strategically integrate multiple external knowledge sources into the pre-trained language model fine-tuning process for TABFSA. Experiments on the FiQA Task 1 and SemEval 2017 Task 5 datasets show that the knowledge-enabled models systematically improve upon their plain deep learning counterparts, and some outperform the state-of-the-art results reported in terms of aspect sentiment analysis error.