Inteligência de Estado e documentos desclassificados da CIA: uma abordagem de Aprendizado de Máquina

Abstract

State Intelligence is an activity of advising the central decision-making power of National States. The production of knowledge of the State Intelligence activity generates documents classified in degrees of secrecy and temporality. After the defined time, the documents are made available for public consultation. Thus, the purpose of this research was to retrieve the main terms/themes/topics followed by the North American External State Intelligence in a corpus related to economic issues at the international scope. The premise is that the results found express international interests of the North American State, since the State Intelligence exists due to the decision maker in the last resort. The analytical step applied to natural language processing techniques and machine learning to perform documentary analysis in order to discover and discuss the most relevant subjects, locations and entities that were the object of observation by the United States Intelligence, if and how they related to Brazil. The Python language was the means used to develop algorithms capable of performing the recovery and characterization of the document’s corpora, through which term counting, word-emdeddings and clustering were applied, culminating in the topic modeling. Six final topics were defined that addressed themes related to the USSR, military, political and economic power, oil and its products, grains, transport and services, aviation and navy, China, Brazil, Africa, Arab’s world and Europe.