In the ideal world of scholarly publishing, every researcher would have a single, permanent digital fingerprint. In reality, the scientific record is a messy collection of strings. A single person might appear as "J. Smith," "John Smith," and "John A. Smith" across different journals—a problem known as synonymy. Conversely, dozens of different individuals may share the exact same name—a problem known as polysemy.
Resolving these identity conflicts is not just a matter of administrative tidiness. As Gabriel Kreiman and John H. R. Maunsell argue in Nine Criteria for a Measure of Scientific Output, the lack of universal identifiers makes it nearly impossible to fairly evaluate a scientist's impact. Without robust Author Name Disambiguation (AND), the metrics we use to fund and promote researchers become unreliable.
The Cost of Identity Fragmentation
When a researcher’s work is split across multiple profiles, their cumulative impact is obscured. Simone Scarcella and colleagues, in their Correction to: Concomitant robot-assisted laparoscopic surgeries for upper and lower urinary tract malignancies: a comprehensive literature review, highlight how even minor formatting errors hinder the ability of the community to accurately attribute contributions.
This fragmentation has bizarre consequences for large-scale data analysis. In the study GeoDBLP: Geo-Tagging DBLP for Mining the Sociology of Computer Science, researchers found that poor disambiguation creates "teleporting researchers"—artificial trajectories where two different people with the same name are merged, making it look like a single individual is moving between institutions in impossible ways. Furthermore, Xiaojun Wan notes in x-index: a fantastic new indicator for quantifying a scientist's scientific impact that name ambiguity distorts impact rankings, particularly for researchers with common surnames.
How Disambiguation Works
To move from "strings" to "entities," researchers use a combination of persistent identifiers and algorithmic inference.
Persistent Identifiers and Metadata
The most direct solution is the adoption of unique IDs like ORCID. Tin Wee Tan and Asif M. Khan proposed mapping these identifiers through a DNS-based system to ensure accurate citation metrics. Similarly, Christian T. Jacobs developed PyRDM to automate the linking of software and datasets to these persistent researcher profiles.
Algorithmic Inference
Because millions of legacy papers lack unique IDs, we must use algorithms to "guess" which names belong together. Shubhanshu Mishra and Vetle I. Torvik demonstrate in Self-citation is the hallmark of productive authors, of any gender that metadata like institutional affiliations, co-author networks, and research sub-fields can distinguish between homonyms with high precision.
Modern systems like CiteSeerX, detailed by Jian Wu and Kyle Mark Williams, use Random Forests to scale this process across millions of documents. For even larger datasets, like the USPTO patent database, Madian Khabsa and C. Lee Giles combined Random Forests with DBSCAN clustering to resolve 12 million inventor mentions in mere hours.
Cultural and Linguistic Nuance
A major hurdle in this field is the "one-size-fits-all" approach to names. Gilles Louppe and Mateusz Susik argue in Ethnicity sensitive author disambiguation using semi-supervised learning that standard string-matching fails for non-Western names. For example, the high prevalence of shared surnames in Chinese academic journals requires richer, linked metadata—such as project associations—to distinguish between researchers, a challenge addressed by Peng Xu and Haofen Wang in LinkedCJ.
Toward a Unified Literature Graph
The ultimate goal is to move beyond isolated databases toward a unified "Knowledge Graph." Dirk Groeneveld and Waleed Ammar describe this in Construction of the Literature Graph in Semantic Scholar, where entity linking transforms millions of papers into a structured network of authors and institutions. Similarly, Jian Xu and Min Song built the PubMed Knowledge Graph by integrating ORCID and NIH data to track career trajectories across time.
To ensure these efforts are reproducible, HaeJin Lee and Jana Diesner have proposed SoDA Cards, a standardized reporting framework that makes the methodological choices behind author disambiguation transparent. By treating identity resolution as a core scientific infrastructure, the field is moving toward a future where every contribution is correctly attributed, regardless of how a name is spelled.