Connected Papers

Technical Documentation

Connected Papers does not expose a public API or SDK at this time. The tool is entirely web-based and works as follows:

• Data source: Powered by the Semantic Scholar Open Research Corpus, which indexes hundreds of millions of academic papers.
• Graph construction: Given a seed paper (identified by title, DOI, or URL), the backend computes a similarity score between papers based on co-citation and bibliographic coupling — papers that frequently appear together in reference lists are considered more similar, regardless of whether they directly cite each other.
• Node attributes: Each node encodes metadata including title, authors, year, citation count, and abstract snippet.
• Edge weights: Edges reflect the degree of similarity; stronger similarity = closer proximity and thicker edge in the rendered graph.
• Graph layouts: Force-directed layout algorithm positions nodes so that highly similar papers cluster together visually.
• Prior Works / Derivative Works panels: Separate views surface foundational papers that influenced the seed, and later papers that built upon it.
• Integration: Papers can be saved to personal libraries or opened directly in Semantic Scholar or external DOI resolvers.