When selecting conferences for csindexbr, we found a relevant correlation between the h5-index and the number of papers accepted in CS conferences.
Basically, h5-index is the largest number h such that h articles published in the last five editions of a conference have at least h citations each.
For example, see the following scatterplot, considering 166 conferences; h5-index is as computed by Google Scholar (in June, 2018).
These conferences cover 18 sub-areas. Although most CS sub-areas are included in the list, we acknoledge that some important sub-areas are still not covered by csindexbr (e.g. Operating Systems and Computer Graphics).
Our theory is that if a conference accepts many papers (hundreds of papers, for example) it produces many citations; some of these citations will refer to papers accepted in previous editions of the same conference. Naturally, some of these papers receive more citations and therefore increase the conference's h-index (or h5-index).
For example, there are two conferences with h5-index > 100 in our dataset (both accepting more than 600 papers):
1.CVPR (Computer Vision): 783 papers accepted in 2017; h5-index= 158
2.NIPS (Artificial Intelligence): 678 papers accepted in 2017; h5-index= 101
Check also the next 13 conferences with the highest h5-index (as we can see, 6 conferences accept more than 400 papers):
3. ECCV (Computer Vision): 415 papers; h5-index=98
4. ICML (Artificial Intelligence): 433 papers; h5-index=91
5. ICCV (Computer Vision): 621 papers; h5-index=89
6. CHI (Human-Computer Interaction): 600 papers; h5-index= 85
7. INFOCOM (Networks): 292 papers; h5-index= 80
8. WWW (Web): 164 papers; h5-index= 77
9. KDD (Data Mining): 130 papers; h5-index= 73
10. VLDB (Databases): 133 papers; h5-index= 73
11. ICRA (Robotics): 939 papers; h5-index= 71
12. CCS (Security): 151 papers; h5-index= 71
13. IROS (Robotics): 970 papers; h5-index= 68
14. ICSE (Software Engineering): 68 papers; h5-index= 68
15. IEEE S&P (Security): 60 papers; h5-index= 68
Then, we decided to compute the normalized h5-index, defined as:
normalized-h5-index= h5-index / papers-accepted-2017
Ideally, we should normalize by the total number of papers accepted in the last five years; however, we only collected statistics about papers accepted in 2017. Despite that, we think the number of accepted papers in a conference does not change a lot, among recent editions.
(It is also important to mention that the idea of normalizing h-indexes is not new; for example, it is used in this paper from 2007).
The following chart shows the top-15 conferences, with normalized h5-index ≥ 1
Some observations about these 15 conferences:
- They accepted a small number of papers in 2017. For example, CLOUD accepted 20 papers; SIGCOMM accepted 36; and SPLC accepted 15 papers.
- NSDI is the conference which accepted the highest number of papers (46 papers).
- SIGCOMM is also the conference with the highest h5-index (67), followed by NSDI (62).
- SPLC is the conference with the lowest h5-index (25), followed by MODELS (26).
The next chart shows the next 16 conferences, with normalized h5-index ≥1.
