Aplicação de técnicas de mineração de textos para classificação de documentos : um estudo da automatização da triagem de denúncias na CGU

The Office of the Comptroller General (CGU) is the agency of the Federal Government in charge of assisting the President of the Republic in matters related to internal control activities, public audits, corrective and disciplinary measures, corruption prevention and combating and coordinating ombudsman’s activities. Through a complaints link of the CGU site, citizens have access to a form to file their complaints. These complaints must be screened and delivered to the coordination of CGU by subject. Nowadays the complaints screening is done manually and they are delivered to one of the 91 coordinating units of CGU. This large amount of categories is more complex in automatic text classification. Considering the complaints storage on the database now and the arrival of new complaints, combined with the time spent on manual sorting, the timely analysis of the reported occurrences it becomes increasingly difficult. This context can cause financial losses to Federal Public Administration as well as discouraging the use of the channel by the citizen. Complaints registered origins are municipalities present in all Brazilian states, generating a great impact on the entire national territory. This research intends to develop a proof of concept for an automatic model of complaints screening, using text mining. The best results were achieved using ranking based on the Huffman Tree algorithm. This proof of concept demonstrated the feasibility of automatical sorting without the loss of quality compared to manual sorting.