Classificação de área científica de artigos utilizando a rede de citações e grafos convolucionais

Abstract

The search for scientific papers is usually done by keywords, but there is no guarantee that these words are exclusive to a single topic. This imprecision tends to generate a large volume of documents in the search processes, thus making the search process difficult. In this sense, researchers made efforts to include in this search the citations that these documents made to other similar and/or related documents. The addition of this information tends to retrieve documents that are more accurate and focused on the topic of interest. From a computational point of view this problem can be seen as a graph node classification, where the articles are the nodes and the citations are the edges. One of the most prominent methodologies in this area is graph convolution, inspired by the success that convolutions have had with images and audio. Following this line, this work uses a convolutional graph model to classify the areas of a scientific article using as a criterion its citation network and a counter of keywords mentioned in its text. The results obtained were satisfactory.