Virtual Investment Researcher

SEWA will gather relevant information sources through a combination of searches on the company, relevant websites, and public and internal databases. This information may be in textual or numerical form represented in HTML, Word, PDF, Excel or SQL formats. Semi-automating this process requires the development of a system that can take a company name as input, search the relevant websites, intranets, and databases, handling the different formats, and return a superset of potentially relevant sources from which the researcher can quickly select. In order to exploit techniques from the field of information retrieval, such as query expansion and relevance feedback to ensure all sources are found efficiently, the research team will need to develop a bespoke solution built on an open source platform. To achieve the full functionality required for the application, this solution will need to go beyond the currently available technology by, for example, being able to accurately index and extract text sentences and paragraphs from PDF documents.

Text analytics or information extraction will require the system to learn contextual patterns encoding relevant types of information on the basis of the links between text regions and report content. To customise such technology to the company report application will require the researchers to deploy either open source platforms or flexible commercial toolkits, for example for named entity recognition or relation extraction and the formatting of output in XML. The researchers provide input to the application developers, for example to optimise the reliability and accuracy of information aggregation.

Automation of text generation will be undertaken using the structured XML content extracted. The university research team will develop and apply appropriate algorithms which are able to generate natural language, and produce pre-defined text according to rules set up by the business team. The researchers will also provide input to the software development team on the integration of the selected algorithms and toolkits into the CUI.

This project is led by All Street Research Ltd, with the intent of developing a working prototype of their proposed autonomous investment research system, SEWA. We will make fundamental contributions to the underpinning machine learning framework, which is vital for delivering the prototype. A successful outcome will have a huge positive impact on the company, enabling them to move into commercial deployment of their product, raise investment, and enable them to substantially increase their subscription revenue base.

Longer term, there will be significant impact on access to investment for SMEs, as SEWA will enable more comprehensive investment reports to be produced much more rapidly, and be much more widely available at an affordable price to both institutional and individual investors.

In addition, this project will help cement the relationship between All Street and the University of Cambridge Computer Laboratory, which is one that we believe could lead to longer term collaboration.