The dataset contains all paper titles, authors and their affiliations from the years

• ICML: 2017-2023
• NeurIPS: 2006-2022
• ICLR: 2018-2023 (except 2020)

The earliest years are always the years in which the respective conference introduced the web interface which this script is compatible with.

Conference,Year,Title,Author,Affiliation
NeurIPS,2006,Attentional Processing on a Spike-Based VLSI Neural Network,Yingxue Wang,"Swiss Federal Institute of Technology, Zurich"
NeurIPS,2006,Attentional Processing on a Spike-Based VLSI Neural Network,Rodney J Douglas,Institute of Neuroinformatics
NeurIPS,2006,Attentional Processing on a Spike-Based VLSI Neural Network,Shih-Chii Liu,"Institute for Neuroinformatics, University of Zurich and ETH Zurich"
NeurIPS,2006,Multi-Task Feature Learning,Andreas Argyriou,Ecole Centrale de Paris
# ...

In 2020 the corona virus spread over the world and forced the conferences to adopt a new virtual format. ICLR decided to announce the conference schedule via a separate page for just this one year. For 2021 their schedule includes poster sessions for all papers again as usual. This alternate schedule page is not covered by the scraping script which is why there are no papers for ICLR2020.

The first and definitely correct option is to re-scrape the whole dataset as in the following example.

python scrape.py 2006-2021

A faster alternative is just scraping the new data and appending it to the CSV file.

python scrape.py --output update.csv 2019-2021
cat update.csv >> papers.csv

The file is sorted by year, so appending at the end keeps the order in tact. However, you need to take care that you do not end up with duplicate entries. Let's say that the current file contains all papers until 2019 but when the file was created, only ICLR had happened yet. If you then later scrape 2019 again to add the other conferences as above, you would get the ICLR papers twice.