Generating narratives from financial data using active learning

点击翻译按钮获取中文摘要

One of the major responsibilities of financial and business analysts is to generate narrative reports summarizing**business trends based on time series data. As it stands, this analysis is typically done manually, requiring**analysts to pour over huge amounts of data looking to understand trends, and performing ad hoc analyses to**establish relationships between data streams. As such, it is currently a high touch process and even with the**many analysts currently employed, important trends can go undetected for long periods of time. As in other**fields, a push toward automation has the potential to increase efficiency in the financial and business analysis**sector, increasing the speed of analysis, decreasing error rates, and freeing analysts from low-level tasks to**allow them to focus on high-level synthesis. One of the challenges in automating financial and business**analysis tasks is that analysts tend to rely on years of experience and domain specific knowledge to achieve**good results. Thus they rely on a "Gestalt" to lead their analysis, which is difficult to replicate using a pure**rule-based system. As such, Unilever Canada aims to develop a neural network based system for generating**narrative reports in natural language from numeric time series data, that will build on recent progress in text**generation from AI subfields such as machine translation, speech-to-text, image captioning and question**answering.