A question answering system for financial economic domain

• Grant number: 20/05895-0
• Support Opportunities: Research Grants - Innovative Research in Small Business - PIPE
• Start date: May 01, 2021
• End date: January 31, 2022
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Principal Investigator: Valdirene Fontanette
• Grantee: Valdirene Fontanette
• Company: Itera Inovação e Desenvolvimento Tecnológico Ltda. - ME
• CNAE: Desenvolvimento e licenciamento de programas de computador não-customizáveis ; Suporte técnico, manutenção e outros serviços em tecnologia da informação ; Tratamento de dados, provedores de serviços de aplicação e serviços de hospedagem na internet
• City: São Carlos
• Associated researchers: Solange Oliveira Rezende
• Associated research grant(s): 23/09956-1 - Alice Balanço Q&A: question and answer system for financial services institution, AP.PIPE
• Associated scholarship(s): 21/02827-6 - A question answering system for the financial economic domain, BP.TT; 21/03210-2 - A question answering system for the financial economic domain, BP.TT; 21/03211-9 - A question answering system for the financial economic domain, BP.TT

Abstract

One of the services in Itera's portfolio is Extractor IA, a solution that helps customers in the segment of financial services institutions to automate part of the back office of risk analysis operations by automating the process of extracting balance sheet data. These institutions use this data to calculate the score of companies that demand their financial credit offers, indicating how reliable or risky it is to provide credit for a given company. With the automatic extraction of the Extractor IA service, manual work is almost extinct, being kept in a minority way for validations and error corrections susceptible to any computer system. For financial institutions to calculate the score of companies that demand for their credit offers, they need to use not only data from the balance sheet, but also need external information obtained from searching in communication vehicles such as newspapers and online magazines that mention the names of these companies under aspects related to corruption and investments, whether they are about acquisition, new products or innovation. These vehicles are consulted by credit analysts in search for information that may, in some way, increase or decrease the risk in granting credit to the applicant companies. In this context, a question answering system could meet this need for information quickly and objectively, automatically providing answers to questions about these companies. Thus, focusing on the provision of services provided by Itera's Extractor IA and its large coverage in terms of back office automation for the risk analysis operation of these institutions, Itera believes that extractive approach techniques can be applied to generate responses from automatically using news from online magazines and newspapers as a source of knowledge, enabling the construction of a question answering system for the economic and financial domain. Therefore, the objective of this project is to evaluate the feasibility of applying an extractive approach via the development of a question answering system about companies using digital media news as a source of knowledge. It is believed that, commercially, the main result of this research proposal is to validate the use of the extractive approach as a possibility to increase the scope of use of the Extractor IA service as the most valuable offer from Itera for clients in the segment of financial institutions. It is also believed that, technically, we will result in the development of new technical capabilities of the Itera team, researching, testing, implementing and validating the extractive approach to generate answers, enabling them, including the possibility of applying such an approach to other domain problems, in addition to the one covered in this project. Proving the feasibility of using the extractive approach for question answering surveyed in phase I, it is kept in mind that new investments must be made to improve the prototype that will be built, researching and comparing the extractive approach with other possible approaches to solve the problem of the question answering domain, investigating new and better techniques for visualizing the generated data, enabling the prototype to process large volumes of data in a big data environment in a scalable cloud software architecture. (AU)