Financial Causality Extraction using Graph Neural Networks

Abstract

In this paper, we propose a method for extracting financial causal knowledge from multilingual text data. In the financial field, fund managers and financial analysts need causal knowledge in their work. Existing language processing techniques are very effective in extracting causal knowledge recognized by humans, but existing methods have two major problems. First, multilingual causality extraction has not been established so far. Second, the technology for extracting complex causal structures, such as nested causal knowledge, is insufficient. To solve these problems, we propose a method to extract nested causal knowledge based on clues (because, due to, etc.) and syntactic information. As a result of evaluating the proposed financial causal knowledge extraction method with multilingual text data in the financial field, it was demonstrated that the proposed model outperforms existing models.