Classificação Automática de Produções Científicas em Inteligência Artificial Utilizando Processamento de Linguagem Natural.

Abstract

The analysis of scientific communities is an important topic for understanding sci entific scenarios from different perspectives. This subject gains strength with the de velopment of digital libraries. In this context, there are different virtual repositories that provide bibliographic data of scientific productions and information about authors and venues. In Brazil, the main scientific database is the Lattes Platform, which has mil lions of researchers’ CVs. However, the platform lacks automated tools to analyze data and faces problems related to manual filling of free text fields. Thus, the task of ana lyzing the volume of publications by subject can be a difficult task. Machine learning models applied with natural language processing techniques are being shown to be a useful alternative for text classification. In this work, models were developed to classify subarea and specialty of cientific works, based only on the title. As a case study, the subarea of Artificial Intelligence and its specialties was used. In this way, datasets were created from extraction of The DBLP Computer Science Bibliography (DBLP) data for developing of two models: one to classify whether a given title of Computer Science is related to the subarea of AI; another to classify, among seven categories, which is the specialty of AI. The models achieved accuracy of 93% and 71%, respectively.