| paper | d4-conf-ijcai-2017-0824 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | high |
| authors would recognise it | yes |
Theorem 1 gives a worst-case price-of-diversity bound, an efficiency-loss result rather than an algorithm, complexity bound, or hardness result. The paper's greedy approximation claim is unnumbered, so the plausible mass-splitting panel model cannot supply the missing named computational anchor.
fails bit a — no named computational result to mirror
The proposed mass model covers the diverse matching objective and right-constrained price-of-diversity analysis; it leaves the paper's greedy guarantee and empirical results outside the stated question.
The strongest honest case is narrow: mirror the paper’s diverse \(b\)-matching problem for a high-volume market with repeated, standardized profiles. The paper has no numbered complexity-class result. Its sole numbered theorem, Theorem 1, is a quantitative optimization result, so I use it as the anchor and do not treat the paper’s unnumbered NP-hardness claim or greedy guarantee as anchors.
Theorem 1, proved in this paper, states that the worst-case price of diversity for a right-constrained diverse matching is
\[
\frac{1}{N}\sum_{l=1}^{N}
\frac{z_l}{1+\sqrt{R^- - 1}\sqrt{z_l^2-1}},
\qquad
z_l=\frac{\sum_j w_{jl}}{\min_j w_{jl}},
\]
where \(N\) is the number of right-side nodes, \(R^-\) is their lower cardinality bound, and \(w_{jl}\) is the weight of the edge from left-side node \(j\) to right-side node \(l\). The paper presents this as a worst-case price-of-diversity bound.
A plausible regime is a large review-assignment market using a standardized topic rubric: many submissions and reviewers, but relatively few exact profiles. A paper type records its complete topic profile and assignment parameters; a reviewer type records expertise, diversity cluster, load limits, and the scores induced against each paper type. Mass is the fraction of submissions of each type and the reviewer supply per unit of submission population. For example, a type might account for \(8\%\) of submissions, while a reviewer type supplies \(0.12\) reviewers per submission. The profiles must determine the same scores and constraints for everyone of that type; coarse grouping that hides meaningful score differences would change the problem.
I would call the mirror Continuous Diverse Panel Matching. Let \(Q\) be the paper types, with masses \(\mu_q\) summing to \(1\), and \(H\) the reviewer types, with supply intensities \(\nu_h\). Let \(\kappa(h)\) be reviewer type \(h\)’s diversity cluster, and \(w_{hq}\) its match weight with paper type \(q\). Fix a panel size \(r\). A panel configuration is a vector \(a\in\mathbb Z_{\ge 0}^{H}\) with \(\sum_h a_h=r\); its entries count reviewers of each type in the panel. Let \(y_{q,a}\) be the mass of type-\(q\) papers assigned configuration \(a\). Require
\[
\sum_a y_{q,a}=\mu_q
\quad\text{and}\quad
\ell_h\nu_h\leq\sum_{q,a}a_h y_{q,a}\leq u_h\nu_h
\]
for each reviewer type \(h\), where \(\ell_h,u_h\) are its per-reviewer load bounds. Each individual panel remains an integral bundle; \(y\) allows a type’s population to be split across different bundles. The objective is to minimize
\[
\sum_{q,a}y_{q,a}
\sum_k\left(\sum_{h:\kappa(h)=k}a_h w_{hq}\right)^2.
\]
The solution is the minimizing allocation \(y\), together with its price of diversity relative to the paper’s linear WBM objective on the same feasible allocations. The specific question prompted by Theorem 1 is whether its worst-case price bound has a valid, tight high-multiplicity counterpart under this mass model and the appropriate scaling of reviewer supply.
For fixed \(r\), this mirror looks Class A: the number of configurations per paper type is \(\binom{|H|+r-1}{r}\), so the configuration LP has polynomially many variables in the type support when panel size is fixed. With variable \(r\), that formulation can grow exponentially, and bundle selection or pricing may retain hardness. The paper’s cited NP-hardness for constrained WBM comes from Chen et al.; the paper’s own general NP-hardness statement is unnumbered, so I do not use either as a named anchor here. This case covers the diverse matching objective and the right-constrained price-of-diversity analysis, not every result in the paper. Further questions include whether the bound extends under binding reviewer capacities and whether variable-\(r\) pricing is tractable.
The weak point is the regime assumption: exact types include the full score behavior and load parameters, so real review markets may have nearly as many types as people. Standardized profiles and score rules make repeated types plausible in some markets, but the paper’s small, idiosyncratic datasets do not establish that this scaling regime is common.
Theorem 1 is a weak anchor for ChoCo: it gives a worst-case price-of-diversity bound, not a complexity or algorithmic result. Once papers are grouped into types, its per-paper expression can be averaged over the type masses. Reviewer supplies change the weights and their scaling, but that produces a continuous version of the bound, not a new computational question of the kind this programme is meant to chart. A theorem about whether that bound is tight in the limit could be interesting, but it would be an analytic question.
There is a stronger possible mirror than the proponent’s bound question: assign each paper type a distribution over integral reviewer panels, and couple those assignments through reviewer-type capacity constraints. That is a sensible high-multiplicity model for a market with standardized profiles. It preserves the paper’s diversity objective and its concern with reviewer loads. I can’t dismiss it on grounds of individual identity or fractional panels: the panels stay integral, while mass is split across them, just as the programme permits.
The best negative case is that this mirror may have little of the paper’s computational contribution left to study. The paper’s named result is not computational, and its algorithmic claims are not formalized as numbered complexity theorems. In the proposed model, the main extra step is to enumerate panel configurations and use their costs as columns in a linear program. With fixed panel size, that configuration space is polynomial in the number of reviewer types; with variable panel size, the difficult part is choosing a panel for one paper type. That difficulty is still worth classifying, but the continuous population itself may add little: it relaxes how often each panel configuration is used, while the local panel-selection problem remains.
That case against is not strong enough to support the universal claim. Standardized paper and reviewer profiles are plausible, and the configuration model is a genuine continuous-population question. The proponent’s specific Theorem 1 anchor is poorly matched to ChoCo’s computational focus, but the better optimization mirror survives. I could argue that it is a routine high-multiplicity reformulation; I cannot honestly argue that no scenario makes it worth studying.
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.