CiteSieve is a lightweight command-line tool that analyzes citations of a given academic paper across OpenAlex and Semantic Scholar. It helps you identify which papers actually use the idea (vs. just mentioning or reviewing it). Perfect for meta-research, survey curation, or evaluating research influence.

Google Scholar numbers are high but opaque (merged versions, many doc types). CiteSieve gives you:

• Transparent sources (OpenAlex + Semantic Scholar)
• Reproducible filters (type and title-based)
• A usage estimate via lightweight textual signals

• Combines OpenAlex and Semantic Scholar for wide citation coverage.
• Filters out reviews, books, surveys, and non-research document types.
• Flags likely usage citations via keyword heuristics (e.g., "decoupled propagation", "A^K X", etc.).
• Produces transparent .csv and .txt outputs for reproducibility.
• Fully configurable via YAML.

git clone https://github.com/<your-user>/CiteSieve.git
cd CiteSieve
python -m venv .venv && source .venv/bin/activate
pip install -e .

Dependencies: Python 3.8+, requests, tqdm, pyyaml

citesieve --title "Simplifying Graph Convolutional Networks" --out sgc_sieve

citesieve --openalex W2916106175 --openalex W2964124573 --out sgc_sieve

citesieve --title "Simplifying Graph Convolutional Networks" --year-min 2019 --year-max 2025 --out sgc_2019_2025

citesieve --title "Simplifying Graph Convolutional Networks" --config my_custom.yml --out sgc_custom

Each run generates two files in the working directory:

Example console output:

[WORK IDS]
- W2964124573 | year=2019 | cited_by=351 | Simplifying Graph Convolutional Networks
  https://openalex.org/W2964124573
[INFO] OpenAlex citers: 1536
[INFO] Semantic Scholar citers: 3486
[INFO] Union unique-by-title: 3744
[STATS]
Removed by TYPE: 84  (book: 20, review: 42, other: 22)
Removed by TITLE: 48  (survey: 26, tutorial: 12, benchmark: 10)
Kept (final):    3612
Likely usage:    278
Titles: sgc_sieve.titles.txt
CSV   : sgc_sieve.filtered.csv

1. Type filters exclude books, theses, reports, etc.
2. Title filters exclude “survey”, “review”, “overview”, “tutorial”, etc.
3. Usage hints flag papers with textual evidence of methodological adoption, e.g.:
   • precompute features
   • decoupled propagation
   • A^K X
   • linearized GNN
   • post-propagation

All regex patterns are configurable in the YAML file.

Default config (auto-loaded unless you pass --config):

filters:
  exclude_types:
    - book
    - book-chapter
    - book-part
    - book-section
    - book-series
    - book-track
    - monograph
    - reference-entry
    - encyclopedia-entry
    - edited-book
    - report
    - report-component
    - dissertation
    - other
    - standard

  title_patterns:
    survey:            "\bsurvey\b"
    review:            "\breview\b|\bliterature review\b|\bcomprehensive review\b|\bsystematic review\b|\bscoping review\b|\bmini-review\b|\bmeta[-\s]?analysis\b"
    state_of_the_art:  "\bstate[-\s]?of[-\s]?the[-\s]?art\b"
    overview:          "\boverview\b"
    tutorial:          "\btutorial\b"
    benchmark:         "\bbenchmark(s)?\b"
    bibliometric:      "\bbibliometric\b"
    position:          "\bposition paper\b"
    editorial:         "\beditorial\b"
    handbook:          "\bhandbook\b"
    encyclopedia:      "\bencyclopedia\b"
    book:              "\bbook\b"

usage_hints:
  - "(precomput|propagat|diffus)\s+(feature|embedding|representation)"
  - "\bdecoupl(ed|e)\s+(propagation|message\s*passing)\b"
  - "\ba\^?k\s*x\b"
  - "\blinear(ized)?\s+gnn\b"
  - "\bpost[-\s]?propagation\b"

http:
  user_agent: "citesieve/0.1 (contact: [email protected])"

• Run once without --year-min/--year-max to get total coverage.
• Inspect filtered.csv in Excel or pandas to slice results (df[df.usage_flag=="yes"]).
• Adjust regex in config for your field (e.g., “transformer”, “fine-tuning”).
• You can skip S2 or OpenAlex by modifying code paths in cli.py (advanced users).

• Google Scholar is not used (no API; scraping avoided).
• APIs have rate limits — large citation lists may take a few minutes.
• Usage heuristics are approximate; you can fine-tune or disable them.
• Titles with typos or alternative names may need manual merging.

pytest tests/
black src/

To add features, open PRs or issues on GitHub. Contributions are welcome!

# 1. Activate your environment
source .venv/bin/activate

# 2. Run full analysis
citesieve --title "Simplifying Graph Convolutional Networks" --year-min 2019 --year-max 2025 --out sgc_full

# 3. Check outputs
head sgc_full.titles.txt
open sgc_full.filtered.csv  # or use pandas

Maintainer: Mohammad Dindoost
Email: [email protected]
Version: 0.1-pre