Document Representation for Text Analytics in Finance

The automated analysis of unstructured data that is directly or indirectly relevant to developments on financial markets has attracted attention from researchers and practitioners alike. Recent advances in natural language processing enable a richer representation of textual data with respect to semantical and syntactical characteristics. Specifically, distributed representations of words and documents, commonly referred to as embeddings, are a promising alternative. Consequently, this paper investigates the utilization of these approaches for text analytics in finance. To this end, we synthesize traditional and more recent text representation techniques into a coherent framework and provide explanations of the illustrated methods. Building on this distinction, we systematically analyze the hitherto usage of these methods in the financial domain. The results indicate a surprisingly rare application of the outlined techniques. It is precisely for this reason that this paper aims to connect both finance and natural language processing research and might therefore be helpful in applying new methods at the intersection of the respective research areas.