Connected Papers: Data Source and How Graphs Are Built
Connected Papers builds similarity graphs of research papers from the Semantic Scholar corpus. How the graph works and how papers get in.
Connected Papers is a web tool that draws a graph of research papers related to one origin paper. It does not search the open web. Its data comes from the Semantic Scholar paper corpus. If you want a paper to appear in Connected Papers, make sure that Semantic Scholar lists it.
Data source
The Connected Papers about page says that its database is connected to the Semantic Scholar Paper Corpus, under the ODC-BY license. The Semantic Scholar FAQ says that Semantic Scholar gets papers from web indexing and from partnerships with journals, indexes and content providers. It names PubMed, arXiv and Springer Nature as examples.
You can start a graph from keywords, a paper title, a DOI, or a paper URL from arXiv, PubMed or Semantic Scholar.
How a graph is built
The about page describes these steps:
- For each graph, Connected Papers analyzes about 50,000 papers.
- It selects the few dozen papers with the strongest connections to the origin paper.
- It places papers by similarity, not by direct citation, so the graph is not a citation tree. Two papers can sit close together even if neither cites the other.
- It draws a force-directed graph, a layout that pulls similar papers together and pushes other papers apart.
The similarity score uses co-citation and bibliographic coupling. Co-citation means that other papers cite both papers together. Bibliographic coupling means that both papers cite the same references. Connected Papers assumes that papers with many shared citations and references cover a related subject.
Prior works and derivative works
Two lists are shown next to the graph:
- Prior works are the papers that the graph papers cite most often. Connected Papers says that these are usually important seminal works in the field.
- Derivative works are papers that cite many of the graph papers. Connected Papers says that these are usually surveys of the field or recent related works.
What it means for site owners
Connected Papers shows academic papers only. A blog post, a product page or other website content cannot appear in a graph. Connected Papers documents no web crawler, so there is no user agent or robots.txt rule for it.
A paper can appear in a graph only if the Semantic Scholar corpus has it. The Semantic Scholar guide explains that search engine. If you want a paper removed from Semantic Scholar, contact Semantic Scholar with the paper URL and the reason. Its FAQ describes this process.