Skip to content

Latest commit

 

History

History
30 lines (24 loc) · 2.56 KB

README.md

File metadata and controls

30 lines (24 loc) · 2.56 KB
Raw

Delving into ChatGPT usage in academic writing through excess vocabulary

Excess words in 2024

Analysis code for the paper Kobak et al. 2024, Delving into ChatGPT usage in academic writing through excess vocabulary.

How to cite:

@article{kobak2024delving,
  title={Delving into {ChatGPT} usage in academic writing through excess vocabulary},
  author={Kobak, Dmitry and Gonz\'alez-M\'arquez, Rita and Horv\'at, Em\H{o}ke-\'Agnes and Lause, Jan},
  journal={arXiv preprint arXiv:2406.07016},
  year={2024}
}

All excess words that we identified from 2013 to 2024 are listed in results/excess_words.csv together with our annotations.

Instructions

  1. All excess frequency analysis and all figures shown in the paper (and provided in the figures/ folder) are produced by the scripts/figures.ipynb Python notebook. This notebook takes as the input a Pickle file with yearly counts of each word (which is too large to be provided here) and several other files with yearly counts of word groups (yearly-counts*). The notebook only takes a minute to run.
  2. These yearly word count files are produced by the scripts/preprocess-and-count.py script which takes a few hours to run and needs a lot of memory. This script takes CSV files with abstract texts as input, performs abstracts cleaning via regular expressions (~1 hour), then runs 
    sklearn.feature_extraction.text.CountVectorizer(binary=True).fit_transform(df.AbstractText.values)
    (~0.5 hours), and then does yearly aggregation.
  3. The input to the scripts/preprocess-and-count.py script are three files:
    • pubmed_landscape_data_2024_v2.zip and pubmed_landscape_abstracts_2024.zip containing PubMed data from the end-of-2023 baseline, available at the repository associated with our Patterns paper "The landscape of biomedical research";
    • and pubmed_daily_updates_2024_v2.zip containg PubMed data from January--June 2024.
  4. This last file is constructed by the scripts/process-daily-updates.ipynb notebook that takes all daily XML files from https://ftp.ncbi.nlm.nih.gov/pubmed/updatefiles/ until 2024-06-30 (from pubmed24n1220.xml.gz to pubmed24n1456.xml.gz) as input. These files have to be previously downloaded from the link above, unzipped, and stored in a directory, from which the scripts/process-daily-updates.ipynb notebook will read, combine, and save as a single dataframe (pubmed_landscape_data_2024_v2.zip).