| paper | A Survey on Rank Aggregation |
| authors | Siyi Wang, Qi Deng, Shiwei Feng, Hong Zhang, Chao Liang |
| venue | IJCAI 2024 |
| filed under | voting · theory |
| judged by | gpt-5.6-luna / xhigh (triple__luna__xhigh__c2r1) |
| judge confidence | high |
| authors would recognise it | yes |
The paper contains no numbered theorem-level assertion about an algorithmic or complexity problem, so bit (a) fails. Its voting material supports natural high-multiplicity mirrors, including continuous Borda aggregation, but those are not named computational results of this paper. Therefore no compliant anchored mirror can produce a green verdict.
fails bit a — no named computational result to mirror
An unanchored mirror would cover BordaCount and similar score-based aggregation over repeated complete rankings, but it would leave the supervised, semi-supervised, probabilistic, interactive, neural, and empirical results untouched.
This paper contains no eligible anchor for a continuous-mirror case. It is a survey, and it states no numbered Theorem, Lemma, Corollary, or Proposition asserting a computational result such as membership in \( \mathrm{P} \), NP-hardness, FPT, or parameterized hardness. Consequently, there are zero anchors and therefore no compliant “one continuous problem per anchor” to formulate.
Its equations and named methods do not change that. Section 2.3 defines rank aggregation; Equations (1)–(10) describe scores, transition rules, or learning objectives; and Section 6 reports empirical comparisons. None is a theorem-level computational claim proved in the paper. The statements that BordaCount is widely used, that MC4 performs well, or that QI-IRA ranks highly are descriptive or experimental, not complexity results.
There is nevertheless a plausible unanchored mirror. For example, a continuous Borda aggregation problem would take items \(U\), a finite catalogue \(T\) of ranking types, and rational masses \(\mu_t\), and compute
\[ S_\mu(u)=\sum_{t\in T}\mu_t\bigl(\operatorname{Len}(t)-\operatorname{rank}_t(u)+1\bigr), \]
returning the items ordered by \(S_\mu\). The regime could be a large population of rankers, users, or retrieval systems drawn from a relatively small set of recurring ranking behaviours. This is a sensible high-multiplicity version of the paper’s BordaCount formulation and is exactly computable in polynomial time.
A more substantive extension would be continuous Kemeny aggregation,
\[ \min_{\pi\in S_U}\sum_{t\in T}\mu_t\,d_K(\pi,t), \]
where \(d_K\) is Kendall distance. It has a clear population interpretation and would plausibly inherit discrete hardness through the high-multiplicity encoding. But Kemeny is not a named computational result of this paper, so it cannot serve as the required anchor.
Thus the strongest honest verdict is: the paper may motivate a continuous rank-aggregation research problem, especially around weighted Borda or expected-disagreement objectives, but it supplies no named computational result on which the requested positive case can be based.
The strongest case against a green verdict is procedural and decisive: there are no eligible anchors. This is a survey, not a theorem paper. Equations (1)–(10) define aggregation rules or learning objectives; Table 2 reports experiments; and the open questions are research directions. Nothing is a numbered theorem, lemma, corollary, or proposition establishing a computational result. Thus there is no paper result to which a compliant continuous question can be attached.
The proposed Borda mirror is also weak as a research anchor. With ranking types \(t\) and rational masses \(\mu_t\), computing
\[ S_\mu(u)=\sum_t \mu_t\bigl(\operatorname{Len}(t)-\operatorname{rank}_t(u)+1\bigr) \]
is merely weighted Borda aggregation. If the support is explicit, it is a direct scan of the input; if all \(m!\) types are admitted, the distribution requires a separate representation, and the apparent difficulty comes from representing that distribution rather than from continuization. The score decomposes item-by-item, so there is no analogue of a substantive pricing or separation frontier. Richer settings involving partial lists, queries, labels, or ranker quality cease to have a fixed ranking type unless that contextual information is added to the type.
That objection is not enough to sustain the universal claim, however. The paper explicitly includes social-choice voting, where many voters with identical complete rankings are a perfectly natural high-multiplicity population. A continuous Borda or Condorcet society is therefore not an artificial reinterpretation. The programme also forbids rejecting a well-posed Class A problem merely because its answer may be easy.
More seriously, the proposed Kemeny extension defeats the claim that no worthwhile mirror exists:
\[ \min_{\pi\in S_U}\sum_t \mu_t d_K(\pi,t). \]
It is a canonical population-level rank-aggregation problem. A discrete profile embeds by setting \(\mu_t=n_t/n\), while repeated ballots naturally collapse into types. It is computational, its discrete hardness transfers, and that transfer is explicitly a legitimate outcome of the programme. Kemeny is not a named result of this paper, so it cannot rescue the proponent’s anchor-based case; but it does show that “no scenario produces a worthwhile continuous mirror” is not defensible.
The honest negative conclusion is therefore narrower: this particular paper supplies no eligible computational anchor, and its most immediate Borda reformulation is too routine to justify a programme project. A universal no-mirror verdict would overreach, because the paper’s voting content supports natural high-multiplicity formulations, with Kemeny as the clearest surviving example.
The adversarial triple: the proponent anchors on up to three named results; the opponent sees that case and must defeat every anchor; the judge decides which case convinced it. These are the pipeline’s own outputs, generated by tools/triple_run.py — no human edited them. The paper’s own text is not reproduced here beyond the quoted statement above.