Utilizando técnicas de aprendizagem de máquina e NLP para extração de informações em licitações do Diário Oficial do Estado do Acre.

Abstract

Information Extraction Systems assist humans in searching for specific information in documents. However, most of these systems do not support documents in the Portable Document Format (PDF), which is widely used. In a PDF document, the text content is mixed with metadata or semi-structured data, which makes it difficult for Natural Language Processing (NLP) algorithms to extract the required information. The Court of Auditors of the State of Acre (TCE-AC) is the supervisory and controlling body of the use of public money and the budget and financial administration of the state of Acre, responsible for analyzing and judging the public accounts of the jurisdictions. Jurisdictions must publish information related to bids both in the TCE-AC bid management system and in the Official Gazette of the State of Acre (DOE), which uses the PDF format. It is the responsibility of the TCE-AC to verify that the bidding information is in both places, thus generating a lot of manual work. In this work, we present a PLN solution with the objective of extracting the DOE acts, automatically categorizing the acts as bidding or not, if so, advanced PLN techniques will be used to process and extract the entities and information from the bidding so that it is possible assist the TCE-AC to verify that the bid is also in the bid management system.