Construction of the Literature Graph in Semantic Scholar

TL;DR

This work presents a deployed, scalable system that constructs a heterogeneous literature graph with over 280 million nodes to organize published scientific data for algorithmic discovery. It reframes graph construction as a set of NLP tasks—metadata extraction, entity extraction, and entity linking—while addressing domain-specific challenges not captured by standard benchmarks. Key contributions include a ScienceParse metadata extractor, a multi-branch entity extraction framework with LM-enhanced representations, KB-grounded entity linking, and a comprehensive set of resources for the research community. The resulting literature graph underpins semantic features in semanticscholar.org and supports advanced queries, author tracking, and future research on citation dynamics, entity coverage, and domain-specific knowledge extraction.

Abstract

We describe a deployed scalable system for organizing published scientific literature into a heterogeneous graph to facilitate algorithmic manipulation and discovery. The resulting literature graph consists of more than 280M nodes, representing papers, authors, entities and various interactions between them (e.g., authorships, citations, entity mentions). We reduce literature graph construction into familiar NLP tasks (e.g., entity extraction and linking), point out research challenges due to differences from standard formulations of these tasks, and report empirical results for each task. The methods described in this paper are used to enable semantic features in www.semanticscholar.org

Figures (1)

• Figure 1: Part of the literature graph.