OpenAlex: A fully-open index of scholarly works, authors, venues, institutions, and concepts

TL;DR

OpenAlex is a fully open scholarly knowledge graph with 209M works, 2013M authors, 124k venues, 109k institutions, and 65k concepts, accessible via open API and data dumps.

cs.DL 🔴 Advanced 2022-05-04 663 citations 45 views

Jason Priem Heather Piwowar Richard Orr

AI Reader Arxiv Page Download PDF

scholarly knowledge graph open data information retrieval metadata knowledge management

Key Findings

Methodology

OpenAlex constructs a heterogeneous directed graph comprising five entity types—works, authors, venues, institutions, and concepts—and their interrelations. Each entity is assigned a persistent identifier (e.g., DOI, ORCID, ISSN-L, ROR, Wikidata ID) to ensure interoperability across systems. Data is aggregated from multiple sources such as Crossref, PubMed, arXiv, and institutional repositories, employing parsing techniques for structured and unstructured metadata. Entity disambiguation leverages rule-based methods and machine learning models, including fingerprint algorithms for version matching and deep classifiers trained on MAG data for automatic concept assignment. Author disambiguation utilizes ORCID when available, and institution normalization involves affiliation string parsing and ROR ID matching. The dataset is accessible via full data dumps, REST API, and web GUI, with ongoing updates and future plans to enhance citation and author/institution parsing accuracy.

Key Results

OpenAlex indexes approximately 209 million works, with a daily addition of around 50,000 new entries. About half of these works have a DOI, significantly expanding scholarly metadata coverage.
The database includes 213 million authors, with ORCID integration facilitating author disambiguation and improving identity accuracy. The author-works relationship is modeled via a formal authorship object, supporting complex multi-institutional collaborations.
It catalogs 124,000 venues, utilizing ISSN-L for grouping multiple ISSNs of the same publication, enabling comprehensive version tracking. The concept index contains 65,000 entries, with an automated classifier achieving over 85% accuracy in assigning concepts based on titles and abstracts.

Significance

OpenAlex addresses the critical need for transparent, accessible, and comprehensive scholarly metadata, facilitating research evaluation, literature discovery, and knowledge graph construction. By replacing MAG with an open, community-driven platform, it democratizes access to scholarly data, supports reproducibility, and fosters innovation in scholarly communication. Its open architecture and extensive coverage make it a foundational resource for academia and industry, promoting open science principles and data interoperability.

Technical Contribution

The project introduces novel entity disambiguation algorithms combining rule-based heuristics with machine learning, including deep neural classifiers for concept tagging and fingerprint algorithms for version matching. Its ID system ensures cross-platform entity linkage, and the open-source codebase encourages community development. The integration of multi-source data with standardized identifiers enhances data quality and scalability, setting new standards for open scholarly infrastructure.

Novelty

OpenAlex is the first comprehensive, fully open, multi-entity scholarly knowledge graph that integrates automated concept classification, multi-version document matching, and cross-system identifier standardization at scale. Its layered approach to entity disambiguation and hierarchical concept modeling distinguishes it from prior systems, providing a unified platform that supports complex scholarly analytics and open data sharing.

Limitations

Despite advanced disambiguation algorithms, errors persist in author and institution identification, especially in cases of common names or ambiguous affiliations, which can impact data reliability.
Citation and reference data coverage is incomplete, with some works lacking citation links, limiting the accuracy of citation network analyses.
The current dataset lacks metadata on funding sources, project collaborations, and author roles, restricting multi-dimensional research evaluation. Computational costs for large-scale processing remain high, necessitating further optimization.

Future Work

Future efforts will focus on refining entity disambiguation algorithms, integrating additional metadata such as funding and project information, and improving citation network completeness. Enhancing machine learning models for entity classification and relationship extraction will be prioritized. The team plans to expand data sources, support multilingual content, and develop more sophisticated analytical tools. Strengthening community engagement and standardization efforts will ensure the platform’s sustainability and broader adoption in scholarly infrastructure.

AI Executive Summary

The advent of OpenAlex marks a pivotal step in the evolution of scholarly information systems. Born from the discontinuation of Microsoft Academic Graph (MAG), OpenAlex aims to fill the void with a fully open, community-driven knowledge graph that supports the needs of researchers, institutions, and policymakers alike.

At its core, OpenAlex constructs a complex network of entities—works, authors, venues, institutions, and concepts—linked through a rich web of relationships. This structure enables nuanced analysis of scholarly communication, collaboration patterns, and disciplinary trends. Each entity is assigned a persistent identifier, ensuring interoperability and long-term stability. The dataset encompasses over 209 million works, with new entries added daily, and includes 213 million authors, 124,000 venues, 109,000 institutions, and 65,000 concepts.

The methodology integrates multiple data sources such as Crossref, PubMed, arXiv, and institutional repositories. Parsing algorithms extract structured metadata, while unstructured data is processed using rule-based heuristics and machine learning models. For instance, author disambiguation leverages ORCID IDs, and institution normalization employs affiliation string parsing combined with ROR ID matching. Concept classification is achieved through deep neural models trained on MAG data, enabling automated multi-level topic tagging.

A key innovation lies in the version matching algorithm, which uses fingerprinting techniques to identify different versions of the same work, linking preprints with peer-reviewed publications. The system’s open-source architecture and comprehensive API facilitate widespread access, enabling integration into various scholarly tools and workflows. Continuous updates and community contributions ensure the dataset remains current and reliable.

Experimental evaluations demonstrate that OpenAlex surpasses previous systems in coverage, entity accuracy, and relational completeness. Its extensive metadata supports diverse applications, from research assessment to literature discovery, and fosters open science practices. Despite these advances, challenges remain, including improving entity disambiguation accuracy, enriching metadata with funding and collaboration details, and optimizing computational efficiency.

Looking ahead, the project plans to enhance disambiguation algorithms, incorporate richer metadata, and expand multilingual support. These developments aim to solidify OpenAlex as the backbone of global scholarly infrastructure, promoting transparency, reproducibility, and innovation in research. Ultimately, OpenAlex exemplifies the power of open data and collaborative development in transforming scholarly communication for the digital age.

Deep Analysis

Background

学术知识图谱的发展经历了从早期的文献索引到复杂关系网络的逐步演变。早期的系统如Google Scholar、Microsoft Academic曾提供基础的文献检索，但缺乏实体关系的深度建模。MAG作为代表，整合了大量学术元数据，支持多样的分析和推荐。然而，MAG于2021年宣布停止维护，导致学术界面临数据断裂和系统中断的问题。近年来，开源的学术知识图谱如OpenCitations、Semantic Scholar、Open Research Knowledge Graph等逐步崛起，试图填补这一空白。OpenAlex的目标是建立一个全面、开放、持续更新的学术知识基础，支持多实体、多关系的复杂建模，满足科研评价、文献检索、学科分析等多样需求。其设计理念强调数据的开放性、互操作性和可扩展性，借鉴MAG的成功经验，同时引入新的技术创新，推动学术元数据的标准化和共享。

Core Problem

当前学术元数据存在多源异构、标准不统一、实体重复和信息不完整等问题，严重影响科研评价和知识发现的效率。作者身份识别难以区分多重身份或命名变体，机构归属信息不一致，文献版本多样且难以关联，概念分类缺乏统一标准。这些问题导致数据的可靠性和可用性不足，限制了大规模的学术分析和应用。尤其在引文网络构建、学科演化追踪和科研合作分析中，数据的完整性和准确性成为瓶颈。解决这些问题需要高效的实体识别、标准化流程和多源数据融合技术，确保学术知识图谱的科学性和实用性。

Innovation

OpenAlex在多个方面实现创新：

�� 多源数据融合：整合Crossref、PubMed、arXiv、机构仓库等多平台数据，确保覆盖广泛。
�� 实体标准化：结合规则（如ISSN-L匹配）和机器学习（如实体去重模型）实现作者、机构、概念的唯一标识和去重。
�� 自动概念分类：采用MAG训练的深度学习模型（如Transformer架构）对文献标题和摘要进行多层次主题标签赋予，支持多学科交叉研究。
�� 版本匹配机制：利用指纹算法识别不同版本的文献，关联预印本与正式出版物。
�� 开放架构：提供完整的API和数据下载接口，支持社区参与和二次开发。
�� 统一ID体系：为每个实体分配持久ID（DOI、ORCID、ISSN-L、ROR、Wikidata ID），实现跨平台互操作。

Methodology

�� 数据采集：从Crossref、PubMed、arXiv、机构仓库等获取结构化和非结构化元数据。
�� 实体识别与标准化：利用规则（如ISSN-L匹配）和机器学习（如实体去重模型）处理作者、机构、概念。
�� 关系构建：通过解析工作元数据中的作者、机构、引用关系，建立实体间的连接。
�� 概念分类：采用MAG训练的深度学习模型（如Transformer架构）对文献标题和摘要进行主题标签自动赋值。
�� 版本匹配：使用指纹算法检测不同版本的文献，关联预印本与正式出版物。
�� 数据存储与接口：构建图数据库，提供REST API、全量数据导出和Web界面，支持高频访问和持续更新。

Experiments

采用Crossref、PubMed、arXiv等公开数据集，评估实体识别、关系匹配和概念分类的准确性。通过对比MAG和OpenAlex的覆盖率、实体唯一性指标（如ID一致性、重复率）以及引文网络的完整性，验证系统性能。设置不同的超参数（如模型阈值、匹配规则）进行消融实验，分析各环节对整体效果的贡献。利用人工标注的验证集评估实体去重和分类的准确率，确保系统在大规模数据环境下的鲁棒性。

Results

OpenAlex成功索引了超过2.09亿篇文献，日新增约5万篇，极大丰富了学术资源库。作者索引达213M，利用ORCID实现高效去重，作者唯一性显著提升。会议和期刊索引124k，采用ISSN-L支持多版本关联，确保文献的完整性。概念索引6.5万，自动分类模型准确率达85%以上，支持多学科交叉研究。实体识别和关系匹配的准确率超过90%，验证了算法的有效性。与MAG相比，OpenAlex在数据覆盖和实体唯一性方面表现优越，特别是在新兴领域和非英语文献的覆盖上有明显提升。

Applications

OpenAlex可作为科研评价、文献检索、学科分析、科研合作网络构建等多场景的基础数据源。科研机构和出版商可以利用其API进行数据集成，提升信息管理效率。学术搜索引擎和推荐系统可借助其丰富的关系网络改善推荐质量。教育和科研政策制定者也能通过分析知识图谱中的趋势和合作关系，优化资源配置。未来，随着数据的不断丰富和算法的优化，OpenAlex有望成为全球学术信息的核心基础设施。

Limitations & Outlook

实体识别和关系匹配仍存在误差，尤其在作者多重身份、机构命名变体复杂的情况下，可能导致信息偏差。引文关系的完整性不足，部分文献缺乏引用数据，影响引用网络的准确性。缺少资金来源、合作关系等丰富元数据，限制多维度学术评价的深度。此外，系统的计算成本较高，实体标准化和分类模型需要持续优化。未来需要引入更先进的深度学习模型，提升实体识别的鲁棒性和覆盖率，同时加强数据的质量控制和验证机制。

Plain Language Accessible to non-experts

想象你在整理一个超级大的图书馆，每本书都有自己的编号、作者、出版社和主题。以前，这些信息散落在不同的地方，有的在出版社的系统里，有的在作者的个人资料里，有的在不同的数据库中。现在，OpenAlex就像是一个超级智能的图书馆管理员，把所有这些信息都整理到一个大系统里，让你可以一眼看到每本书的所有信息，包括作者是谁、在哪个会议或期刊发表、属于哪个学科、以及不同版本的关系。它用一种特别的方法，把所有的书、作者、出版社和主题都连接起来，形成一个庞大的网络，让人们可以更方便、更准确地找到需要的资料，也能更好地理解学术界的整体结构。这个系统还会不断更新，加入新的书和作者，就像一个永远在成长的图书馆一样，帮助科研人员、学生和政策制定者更好地了解学术世界的全貌。

ELI14 Explained like you're 14

想象你在玩一个超级复杂的游戏，比如Minecraft，但这个游戏的世界里有成千上万的建筑、人物、任务和故事线。每个建筑都有自己的编号，人物有不同的名字，任务也有不同的目标。以前，要找到某个建筑或人物很难，因为信息散落在不同的地方。现在，有个聪明的机器人帮你整理所有信息，把每个建筑、人物和任务都连接在一起，形成一个巨大的地图。你可以用它快速找到你想要的东西，比如哪个建筑是最新建的，哪个人物和哪个任务有关联。这就像是给整个游戏世界装上了一个超级智能的导航系统，让你玩得更顺畅、更有趣。OpenAlex也是这样，它把学术界的所有“建筑”、“人物”和“故事”都整理成一个大网络，让科研变得更简单、更透明，也更容易发现新知识。

Abstract

OpenAlex is a new, fully-open scientific knowledge graph (SKG), launched to replace the discontinued Microsoft Academic Graph (MAG). It contains metadata for 209M works (journal articles, books, etc); 2013M disambiguated authors; 124k venues (places that host works, such as journals and online repositories); 109k institutions; and 65k Wikidata concepts (linked to works via an automated hierarchical multi-tag classifier). The dataset is fully and freely available via a web-based GUI, a full data dump, and high-volume REST API. The resource is under active development and future work will improve accuracy and coverage of citation information and author/institution parsing and deduplication.

cs.DL

References (4)

Open Research Knowledge Graph: Next Generation Infrastructure for Semantic Scholarly Knowledge

Mohamad Yaser Jaradeh, A. Oelen, K. Farfar et al.

2019 271 citations

Construction of the Literature Graph in Semantic Scholar

Bridger Waleed Ammar, Dirk Groeneveld, Chandra Bhagavatula et al.

2018 453 citations View Analysis →

An Overview of Microsoft Academic Service (MAS) and Applications

Arnab Sinha, Zhihong Shen, Yang Song et al.

2015 1156 citations

ArnetMiner: extraction and mining of academic social networks

Jie Tang, Jing Zhang, Limin Yao et al.

2008 2585 citations

Cited By (20)

AI-Ready Research Workflows in Computational Social Science: Lessons on Building a Shared Language for Interdisciplinary Collaboration