Comparação de Técnicas de Predição de Links em Sub-redes de Coautoria Formada por Currículos da Plataforma Lattes
The study of Lattes platform allows addressing and analyzing Brazil researchers network which could be useful for defining politics to improve science, technology, and innovation. This work evaluated Lattes Platform coauthorship network. This network evolves over time, which means that new coauthorships will arise in future. Therefore, using link prediction methods in this network would help to identify growing knowledge areas in Brazil. The used technics were Spectral Evolution, wich is new in this context, Common Neighbors, Adamic-Adar and Jaccard. The main goal was to evaluate the link prediction accuracy with different methods at the coauthorship network of Lattes Platform. The Spectral Evolution was worse than the others. Adamic-Adar method presented the best result - 817 times better than the random link prediction.