find_promotion_referees.py

# TODO: Library.get(X) should resolve to Library.get(id=X)
#       Make &   (Document.date > (datetime.now() - timedelta(days=365.25 * 5)) work
#       Make document.year > 2017 work
#       Make Document.in_(library) and Library.contains(document) work using docs() function
#       Remove default behaviour of .limit(10)
#       Document.full() should be searchable, and Document.full == '' SHOULD NOT WORK

"""
Give me:
    - highly cited people;
    - who publish in astrophysics;
    - who haven't published with me in the last 5 years;
    - (optionally) who publish in stellar spectroscopy.
"""

import ads
import numpy as np
from scipy.stats import pareto
from ads import Document, Affiliation, Library
from tqdm import tqdm
from datetime import datetime, timedelta
from peewee import fn

# Score people by authorship position and citation count.
def scores(document):
    x = 1 + np.arange(len(document.author))

    # Use uniform distribution if it's alphabetical from second-author onwards???
    # (e.g., to allow for X Collaboration, AAAA, BBBB)
    # otherwise, use a pareto distribution.
    pdf = pareto(1).pdf(x)
    return (document.citation_count * pdf) / np.sum(pdf)

# First build a library of names of people who I have published with in the last N years.
q = (
    Document.select()
            .where(
                (Document.property == "refereed")
            &   (Document.year.between(2017, 2023))
            &   fn.docs("library/UM9dgiNHTkq4xp7588l7pg")
            )
            .limit(1_000)
)

network = {}
with tqdm() as pb:
    for doc in q:
        for author_norm, author in zip(doc.author_norm, doc.author):
            network.setdefault(author_norm, [])
            network[author_norm].append(author)
        pb.update()

print(f"Found {len(network)} unique authors.")

# Now let's find highly cited people.
limit = 10_000
q = (
    Document.select()
            .where(
                (Document.property == "refereed")
            &   (Document.database == "astronomy")
            &   (
                    (Document.full == "stellar spectroscopy")
                |   (Document.full == "data analysis")
                )
            )
            .order_by(Document.citation_count.desc())
            .limit(limit)
)
    
candidate_scores, candidate_names, candidate_documents = ({}, {}, {})
for document in tqdm(q, total=limit):
    for author, full_name, score in zip(document.author_norm, document.author, scores(document)):
        if author not in network: 
            candidate_scores.setdefault(author, 0)
            candidate_documents.setdefault(author, [])
            candidate_names.setdefault(author, [])

            candidate_scores[author] += score
            candidate_documents[author].append(document)
            candidate_names[author].append(full_name)


# Now sort by score.
ranked_candidates = dict(sorted(candidate_scores.items(), key=lambda item: item[1], reverse=True))
for i, (author, score) in enumerate(ranked_candidates.items(), start=1):
    print(f"{i}: {author} ({set(candidate_names[author])})   {score:.0f}")

    if i >= 1000: break