Current academic search engines only provide a user with a list of popular papers rather than relevant but possibly not so well-known papers. Moreover, it is hard to understand the relationship between papers or the current research landscape from a list. In this talk, we present Etymo, a new graphic search engine which aims to solve these problems.
Google's Pagerank algorithm is based on the hyperlink graph of the web. We, on the contrary, build our knowledge graph from the high-dimensional encoding of the full text papers.
Building a scalable graphic search engine is a challenging task. We need to locate millions of papers as nodes in a three-dimensional space such that similar papers are close to each other and papers are located in the `correct' layers for zooming-in and out. We will discuss our approach to the problem and show how different components of the engine work together. We also show how we can exploit temporal network information to get more accurate results.