Named-Entity Recognition for Portuguese Police Reports

During a criminal investigation several text documents are produced by police officers, creating a deluge of unstructured data obtained from heterogeneous sources. Therefore, identification and recognition of entities, i.e. places, organizations or persons, by a natural language pipeline, with named-entities recognition task, could help police officers to understand and find relevant information in data extracted. We aim to defined a natural language processing pipeline to identify and recognize entities from these police reports, supported by two trained corpus, namely Amazonia and a Portuguese News Corpus. Additionally, we evaluate named-entities recognition systems, focus in Portuguese language, with a dataset produced by the Portuguese police. We then evaluate the performance obtained on the information retrieval process applied to the dataset.