Mentorship in science.

What is in the dataset?

An open dataset connecting academic mentorship relationships with researcher profiles, publication identifiers, semantic representations, and demographic estimates. It enriches Academic Family Tree records with publication information and supports research on mentorship and scientific careers, with strongest coverage and validation in neuroscience and biomedical science.

The 2022 Scientific Data paper explains the construction and validation. The dataset release is dated 2021; its release date and the paper's publication date describe different records.

Start with the relationships

Four CSV tables connect mentorship, researchers, authorship, and publication identifiers. Their column names are case-sensitive.

MentorshipType uses 0 for research assistant, 1 for graduate student, 2 for postdoctoral, 3 for research scientist, and 4 for collaborator. Missing dates or identifiers need explicit treatment in an analysis.

A small first analysis

With Python and pandas installed, download two core files and attach each mentee's research area to their mentorship records. The compressed downloads total about 25 MiB. This example does not require the embedding files.

from urllib.request import urlretrieve
import pandas as pd

base = "https://zenodo.org/api/records/4917086/files"
for name in ("mentorship.csv.gz", "researcher.csv.gz"):
    urlretrieve(f"{base}/{name}/content", name)

relationships = pd.read_csv(
    "mentorship.csv.gz",
    usecols=["CID", "MenteeID", "MentorID", "MentorshipType"],
)
researchers = pd.read_csv(
    "researcher.csv.gz", usecols=["PID", "ResearchArea"]
)
mentees = researchers.rename(columns={"PID": "MenteeID"})
joined = relationships.merge(
    mentees, on="MenteeID", how="left", validate="many_to_one"
)
print(joined["ResearchArea"].value_counts(dropna=False).head())

To attach mentor attributes, rename PID to MentorID and join on that column instead. Check matching rates and missing values before interpreting group differences.

Coverage and responsible interpretation

• The records are crowdsourced and do not represent a complete or uniform census of scientific mentorship. Coverage and validation are strongest in neuroscience and biomedical science.
• Publication matching and semantic representations are estimates. Disambiguation errors and missing publications can affect downstream results.
• Gender and race/ethnicity fields are name-based model estimates, not self-reported identities. They can be noisy and biased, and the categories do not describe the full range of individual identities.
• Use these estimates with uncertainty and sensitivity analyses in aggregate research. They should not be used to decide an individual's identity, eligibility, or opportunity.
• The release contains historical Microsoft Academic Graph identifiers. DOI and PMID provide additional routes to publication metadata.

The paper describes some tables that are not present as separate files in this release. The download list below reflects the actual Zenodo record. Large embedding files are optional for the CSV example.

Cite the dataset

Qing Ke, Lizhen Liang, Ying Ding, Stephen V. David, Daniel E. Acuna (2021). A dataset of mentorship in science with semantic and demographic estimations. Zenodo. https://doi.org/10.5281/zenodo.4917086

Please also cite the data descriptor when using its construction or validation methods. The dataset is shared under Creative Commons Attribution 4.0; credit the creators and identify any changes.