Notícias

Banca de DEFESA: WESLLEY EMMANUEL MARTINS LIMA

Uma banca de DEFESA de DOUTORADO foi cadastrada pelo programa.
DISCENTE: WESLLEY EMMANUEL MARTINS LIMA
DATA: 10/05/2025
HORA: 10:00
LOCAL: Sala de Aula do Núcleo de Computação de Alto Desempenho
TÍTULO: A Framework for Extracting Red Flags in Public Procurement: An Architectural Approach
PALAVRAS-CHAVES: Corruption, fraud detection, attention, PLN, deep learning, text mining, public procurement, government procurement, public tender.
PÁGINAS: 103
GRANDE ÁREA: Ciências Exatas e da Terra
ÁREA: Ciência da Computação
RESUMO:

Procurement fraud generates significant economic and social losses worldwide. Despite the efforts of auditing authorities, manual inspection proves inefficient. Due to the high volume of processes and the time required for each analysis, detection rates are low. Computer models have emerged as a promising way to automate fraud detection and prediction. This thesis introduces a novel framework for automatically extracting red flags in public procurement using deep learning and natural language processing (NLP) techniques. The study addresses three main gaps: the lack of automated solutions for red flag detection, the absence of labeled textual databases, and the scarcity of domain-specific language models. The research presents a framework comprising several models for tasks such as section classification, object classification, named entity recognition, and red flag detection. To address the lack of annotated data, the study compiled multiple datasets with thousands of documents related to public procurement, including more than 420,000 documents in Portuguese. A key contribution is the development of HelBERT, a pre-trained language model for the public procurement domain, trained on a large corpus of Portuguese documents. The findings show that HelBERT and its variants outperform general-purpose and legal-domain models in downstream tasks, such as red flag classification and semantic similarity. For example, HelBERT achieved an F1 score of 94.91% in red flag classification and demonstrated the highest accuracy (97.04%) in object classification using keyphrases.


MEMBROS DA BANCA:
Externo à Instituição - 375.***.***-87 - ANSELMO CARDOSO DE PAIVA - UFMA
Externo à Instituição - 005.***.***-62 - CLEBER ZANCHETTIN - UFPE
Interno - 2167802 - PEDRO DE ALCANTARA DOS SANTOS NETO
Externo à Instituição - 310.***.***-23 - RICARDO AUGUSTO SOUZA FERNANDES - UFSCAR
Presidente - 2061294 - RICARDO DE ANDRADE LIRA RABELO
Interno - 1446435 - VINICIUS PONTE MACHADO
Notícia cadastrada em: 30/04/2025 07:19
SIGAA | Superintendência de Tecnologia da Informação - STI/UFPI - (86) 3215-1124 | © UFRN | sigjb02.ufpi.br.timers 09/05/2025 16:31