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 require higher levels of understanding. Targeted Aspect-based Financial Sentiment Analysis (TABFSA) is an example such tasks, it involves processes like information extraction, specification, domain adaptation. However, little known about the design principles hybrid models leveraging external lexical knowledge. To fill this gap, we define anterior, parallel, posterior knowledge integration propose incorporating multiple sources strategically into fine-tuning process pre-trained transformer for TABFSA. Experiments on Opinion mining Question Answering challenge (FiQA) Task 1 SemEval 2017 5 datasets show that knowledge-enabled systematically improve upon their plain deep learning counterparts, some outperform state-of-the-art results reported terms aspect sentiment analysis error. We discover parallel most effective domain-specific more important according to our ablation analysis.