Link prediction aplicado em grafos de citações de artigos usando o Node2vec

Abstract

Machine Learning applied on graphs is a recent field of study on artificial intelligence area, although it it is already solving difficult problems on daily challenge faced by humans, like DNA decodification, optimization of supply and energy networks, and, in the academic field, it could provide powerful tool to allow the researcher to find correlated papers of high quality. In applications of this problem, academic paper databases are modelated in graphs, and then it is possible to apply graph embedding techniques transform this graph in a d-dimensional vectors that can be used on tradicional machine learning algorithms, then it is made a prediction model based on the original graph features, predicting these missing links is a task name link prediction. However, create those models are no easy task, due the complexity involved on the task. Therefore, this work aims to apply the node2vec algorithm to learn porwerful embeddings of the paper citation network, and create a prediction model to perfom link prediction tasks and find related papers on the network.