Imagine you work for a company that publishes scientific articles created by its customers—primarily researchers and scientists in the field of statistics—whose volume of research has grown so large that it’s not possible for your company to read every paper nor for the researchers to stay on top of everything happening in the field. Your task is to use automated means to find some structure in the collection of scientific articles so that the company can more easily home in on the researchers’ interests and help them research more effectively. To do that, you’ll generate document embeddings, use them to impute a document graph, and execute graph algorithms against this graph in order to generate insights from it.

Now that one of the data scientists at your publishing company was able to extract citations from the full text of the representative text corpus, your task is to generate new vector embeddings based on the citation graph, then cluster the documents using these embeddings to gain insight from the graph structure.

In this liveProject, you’ll implement a Graph Neural Network (GNN). This powerful model will allow you to use the document content from the first liveProject combined with the structure of the citation graph from the second liveProject to build an even more powerful model—one that will predict the sub-field of statistics of each of your customers’ papers.