Conference Partner Index: ranking methodology (CP-I)
The Conference Partner Index is a 0-100 score we compute for every conference in our catalogue. This page publishes the entire method: every factor, every weight, every mapping. Nothing here is proprietary - the point is that you can recompute any score yourself and tell us when we get one wrong.
Algorithm version 1.1 · 5,646 conferences scored · Browse the ranking
Why we built our own index
CCF, ICORE and QUALIS are the catalogues researchers actually cite, and we reproduce all three. But between them they cover only about a fifth of the conferences we track: for roughly four out of five entries, all three rating columns are simply empty. Those are exactly the venues people most need help judging - nobody needs to be told what CVPR is. The index exists to say something defensible about the other 80%.
What the score is not
- It is not a peer-review judgement. We do not read papers and we do not survey researchers.
- It does not compete with CCF, ICORE or QUALIS. Where those catalogues rate a conference, their tier is the single largest input to its score.
- It is not a measure of whether a conference is currently accepting papers. That is a separate filter on the ranking page, deliberately kept out of the score.
- Nobody can pay to change a score, and there is no way to buy a position in any list on this site.
The five factors
| Factor | Weight | What it measures |
|---|---|---|
| Academic recognition | 35% | The highest tier this conference holds in CCF, ICORE or QUALIS. We reproduce those catalogues, we do not second-guess them; a conference listed in none of the three is scored at the neutral baseline, not penalised. |
| Submission selectivity | 20% | Average acceptance rate over the most recent editions for which submitted and accepted paper counts are on record. |
| Editions held | 20% | How many editions the conference has demonstrably held, from the edition number plus the span of years for which we hold records. |
| Community attention | 10% | Distinct researchers on Conference Partner who follow this conference or opened its page in the past 24 months. This measures our own community - mostly computer-science researchers reading in Chinese - which is why we deliberately keep its weight low. |
| Public record completeness | 15% | How much of the public record is on file: official website, full call for papers, acceptance rates, best-paper records, past editions, and whether the latest edition is recent. This measures documentation, not academic quality. It is the one factor a verified organizer can raise directly, and the three biggest items - acceptance rates, past editions and best papers - are exactly the ones only the organizer can supply. |
Missing data is scored at a neutral baseline, never as zero
Most conferences are missing at least one factor. When a factor is unknown we score it at 50 out of 100 - the neutral baseline - so it neither helps nor hurts, and we lower the confidence value published beside the score.
Two alternatives were rejected on purpose. Scoring a missing factor as zero would condemn every conference outside computer science, because CCF and ICORE only cover computing and QUALIS is a Brazilian catalogue - absence from them carries no information about a materials-science conference. Re-normalising over the factors we do have would let a conference with nothing on file except a long history score as highly as one that has actually been assessed by somebody.
This is why a typical score sits between 40 and 50. That is the baseline showing through, not a failing grade - it means we do not know much about that conference yet. Always read the score together with the confidence value.
Exact mappings
Academic recognition
We take the highest tier the conference holds in any of the three catalogues. Averaging them would penalise a conference for being listed in more places - a CCF-A venue also rated ICORE B would score below one listed in CCF alone - which is plainly wrong.
| CCF | A → 100 · B → 82 · C → 66 |
|---|---|
| ICORE | A* → 100 · A → 88 · B → 72 · C → 58 |
| QUALIS | A1 → 100 · A2 → 92 · B1 → 78 · B2 → 68 · B3 → 60 · B4 → 54 · B5 → 48 |
Submission selectivity
We average the acceptance rate over the most recent editions on file (at most five), counting only editions with at least 20 submissions. An acceptance rate of 12% or below scores 100; 70% or above scores 30.
The floor is deliberately not zero. A missing acceptance rate scores the neutral 50, so if publishing a high acceptance rate scored zero, hiding the number would be worth more than being honest about it. With the floor set where it is, and with acceptance data also counting towards the public-record factor, disclosing an unflattering rate is roughly score-neutral.
Editions held
A logarithmic curve reaching 100 at 40 editions: the difference between a first and a fifth edition matters far more than the difference between a 36th and a 40th. We count the edition number on record, the past editions we hold snapshots for, and the span of years for which we hold any records - whichever is largest.
Community attention
Two counts, both de-duplicated per person: registered users following the conference, and registered users who opened its page in the past 24 months. Both curves are logarithmic and saturate at 300 followers and 400 readers respectively.
This is attention within the Conference Partner community, which skews towards computer science and towards researchers reading in Chinese. It is not global popularity, and that is exactly why it is capped at 15% of the score. We do not use the site-wide view counter shown on conference pages for this: that counter has never been reset and does not exclude crawlers.
Public record completeness
Points for each piece of the public record we hold. This measures documentation, not academic quality - and it is the one factor an organizer can raise directly, by claiming the conference and filling in what is missing.
- Official website on file — 10
- Full call for papers on file — 15
- Acceptance rates on file — 30
- Best-paper records on file — 15
- Past editions on file — 20
- Latest edition is within the past 24 months — 10
How to raise a conference's score
Every input is something you can supply. A verified organizer can fill all of it in from the conference page in a few minutes, and scores are recomputed every night. The figures below are the exact point gains under the current algorithm version - they are computed from the same constants the scoring code uses, not written by hand.
| What to add | Score gain | Why it moves the score |
|---|---|---|
| Historical acceptance rates | up to 14.5 points | This is the single biggest lever, because it does two things at once: it is worth 30 points of the public-record factor, and it turns submission selectivity from an unknown - scored at the neutral baseline - into a measured value. A selective conference gains a lot; a conference with a high acceptance rate still gains, never loses. Concretely: a 20% acceptance rate is worth 12.6 points, 40% is worth 7.7, and even 80% is worth 0.5. |
| Past editions | 3.0 points, sometimes more | Worth 20 points of the public-record factor. It often raises the editions-held factor as well: we count the span of years we hold records for, so listing older editions can prove a longer track record than the edition number alone shows. It also powers the conference series page. |
| Best-paper records | 2.3 points | Award-winning papers are the clearest public evidence of what a conference actually publishes, and they are indexed on our best-papers pages where researchers and search engines find them. |
| Full call for papers | 2.3 points | A call for papers of reasonable length, not just a link. Most conferences already have one, which is why it is worth less than the items above. |
Filling in all four typically moves a conference by 10 to 20 points and raises its confidence from around 0.45 to 0.85 or higher - which matters as much as the score itself, because a high-confidence score is the one people and AI assistants will quote.
To do it: open the conference page, claim it as its organizer, and the entry forms for acceptance rates, past editions and best papers appear on that page. Your outstanding items are also listed on the organizer dashboard. Claim a conference
What we deliberately left out
- The view counter shown on conference pages. It has counted every hit since 2013 without ever being reset and without excluding crawlers, so it measures how often a page was fetched, not how many researchers care.
- Whether the conference looks dormant. About half our entries have no edition recorded in the past two years, but that usually reflects our own data going stale rather than the conference stopping. Charging a conference for our maintenance backlog would be unfair, and it would blur two separate questions - how well regarded a conference is, and whether you can still submit to it - into one number that answers neither.
- Citation counts and bibliometric indicators. We do not hold that data, and quoting somebody else's numbers we cannot verify would make the score unfalsifiable.
Known limits
- Comparing a conference in one discipline with one in another is rarely meaningful. Use the per-field ranking on each conference page, which places it among conferences carrying the same research-field tag.
- Acceptance-rate coverage is thin. Fewer than one in ten conferences has usable submitted and accepted counts on file, so for most entries the selectivity factor falls back to the baseline.
- The community factor is biased towards conferences that have been listed here longer and towards the fields our users work in.
- Research-field tags are assigned automatically from conference titles, so a small number of conferences sit in the wrong field ranking. Tell us and we will correct the tag.
How to correct a score
Every score is a function of data you can see on the conference page, so a wrong score means wrong or missing input data. The fastest fix is to claim the conference as its organizer and fill in the missing acceptance rates, past editions and call for papers yourself; scores are recomputed daily. If something else looks wrong, email [email protected] and tell us which conference and which factor.
Stability
A conference's score changes only when its input data changes. Ranks are computed in a single pass over the whole catalogue and stored, so the rank shown on a conference page and the position of that conference in the ranking list can never disagree. Whenever we change a weight or a mapping we increment the algorithm version shown above and recompute everything.