We consider these questions for our data exploration.
About total numbers:
- How many submissions, comments and paper revisions are there on OpenReview?
- How many submissions are accepted? How many rejected/ withdrawn/ unknown?
- How many comments are reviews? What 'type' do the other comments have?
About class distributions:
- What part of submissions are rejected in each venue? And which venues do not have an acceptance decision at all?
- Which venues have relatively many reviews compared to other comment types?
About the distribution of submissions over the venues:
- Which venue has the most submissions?
- Does a venue get more submission each year?
About comments per submission:
- How many comments does a submission get on average?
- Which venues have the most comments? Which has the most comments per submission?
About revision per submission:
- How many revisions are uploaded for a submission on average?
- Which venue has the most revisions per submission?
The data was collected with config_all.json on 03.02.2020. Venues with no submissions are ignored.
All values and graphics are created with statistics_from_json.py.
| Submissions | Comments | Paper Revisions |
|---|---|---|
| 9801 | 63541 | 18580 |
| Unknown | Withdrawn | Accepted | Rejected |
|---|---|---|---|
| 2412 | 784 | 3008 | 3597 |
Definition of types used here:
- Review: 'review' in invitation
- Decision 'decision|acceptance' in invitation
- Plain Comment: 'comment' in invitation
- Other: all other comment notes
These types cover most of the comments. Some notes might be 'misclassified' but we do not really care for this here. We are only interested in a rough overview.
| Plain Comment | Review | Decision | Other |
|---|---|---|---|
| 35195 | 22877 | 5022 | 447 |
We display these distributions as heatmaps for each venue & year.
Submissions with 30 or more comments are ignored as outliers.
Distributions are displayed with violin plots. Top and bottom line mark the highest/ lowest value and the middle line marks the median.
Submissions with 30 or more revisions are ignored as outliers.