An Annotated Corpus of Crime-Related Portuguese Documents for NLP and Machine Learning Processing

Criminal investigations collect and analyze the facts related to a crime, from which investigators can deduce evidence be used in court. It is multidisciplinary applied science, includes interviews, interrogations, collection, preservation of chain custody, other methods techniques investigation. These produce both digital paper documents that have carefully analyzed identify correlations interactions among suspects, places, license plates, entities are mentioned The computerized processing these helping hand criminal investigation, as it allows automatic identification their relations, being some difficult manually. There exists wide set dedicated tools, but they major limitation: unable process reports Portuguese language, an annotated corpus for purpose does not exist. This presents corpus, composed collection anonymized crime-related documents, were extracted official open sources. dataset was produced result exploratory initiative data websites conditioned-access police reports. evaluated mean precision 0.808, recall 0.722, F1-score 0.733 obtained with classification named-entities present documents. employed benchmark Machine Learning (ML) Natural Language Processing (NLP) tools detect correlate Some examples sentence detection, named-entity recognition, terms domain.