Descoberta de tópicos e classificação de textos de licitações promovidas pelos órgãos públicos do estado da Bahia

Abstract

Purchasing and hiring bodies Public contracts are carried out through bidding processes, which aims to fulfill its social functions, respecting principles of legality, impersonality, morality, publicity, public interest, administrative probity, among others [1]. The control of these bidding processes helps the adequate management of public resources. With the aim of improving analysis regarding the tenders, the Public Ministry of the State of Bahia developed the BI – Licitometer. This tool Among its analysis functionalities, it has the ability to group bids by categories through the use of key words. However, the method applied for the characterization of bids has not demonstrated efficiency, generating incorrect classifications and making it difficult the analysis of bids. This work proposes the use of BERTopic to perform topic modeling of tenders carried out in the State of Bahia together with a supervised classification approach and the use of an unsupervised method for classifying tenders relating to advisory/consulting services legal. The objective is to compare the results of the methods classification and select the best one to implement in the BI-Licitometer. The unsupervised classification method used was Zero-Shot Classification and the method supervised was Random Forest Classification.