Transformers for Link Prediction on Graphs

We predict links between nodes of graphs using machine learning algorithms. In this work, we implement the algorithm Transformer that is used in natural language processing tasks to graphs structured data. We evaluate our solution on two datasets. The first dataset is an undirected graph, representing a subset of the collaboration network between authors of articles. The task in the first dataset is to predict the future author collaboration relationship given the previous collaborations. The second dataset is about molecules of protein. The nodes in this dataset represent proteins from 58 different species, with edges indicating biologically meaningful associations between proteins, such as co-expression, physical interaction, homology, or genomic neighborhood. Then we will compare the result of our algorithm with others presented on the common leaderboard for these datasets. The paper accounts for 27 pages, 4 illustrations, 3 tables and appeals for 13 outer resources.