Developing a knowledge inference and reasoning engine to extract meaningful insights from unstructured data using a novel neighbourhood graph approach

点击翻译按钮获取中文摘要

Information overload is a common problem, particularly in data heavy industries like finance. Unstructured text, such as company reports, news, transcripts, emails and memos, contain much valuable information that is often missed by data analysts and key decision makers, for example when assessing risk and investments. Current solutions do not visually display the links in datasets that could help decision makers make better, more informed decisions.Auquan is solving a significant, unmet, global need by developing technology that visually links relevant, but often hidden, insights without overwhelming the user with high volumes of data. Our solution extracts information from both unstructured and structured datasets, leading to better decisions and more productive analysts. Automating information retrieval from unstructured datasets applies to many sectors, not just finance. It is particularly challenging because it requires custom extraction (e.g. differentiating between tables/text) from different data formats (e.g. presentations/PDFs/reports) and domain-specific (e.g. legal language versus financial text) training of NLP algorithms.This project aims to dramatically extend and improve our technology, then undertake pilot evaluations with early adopters to measure/demonstrate that our actionable insights save time and deliver better performance. We also plan to investigate exciting wider use-cases for our technology outside the finance sector including scientific/medical literature.