| paper | d4-conf-ijcai-2017-0345 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | high |
| authors would recognise it | unclear |
The paper states no named, numbered computational result to anchor a continuous-population mirror; its running times and accuracy findings are empirical. Its category distributions concern document contents, while the workers remain a finite set, so the proposed worker-type continuum would be a new project rather than a mirror of a result in this paper.
fails bit a — no named computational result to mirror
The paper's Bayesian model, empirical aggregation and spammer-detection findings, and reported running times have no named computational guarantee to mirror.
I cannot make a result-anchored positive case from this paper. It contains no numbered Theorem, Lemma, Corollary, or Proposition asserting a computational complexity result, so it has no qualifying anchor under your criteria.
The closest candidate is MBCC’s inference model in Section 3: it estimates document-level category proportions \(\Lambda_i\) alongside workers’ confusion matrices. But that is a model formulation, not a named complexity result. And \(\Lambda_i\) makes the *document’s categories* probabilistic; the population of \(K\) workers remains finite. Treating a large crowd as masses of worker types—where a type includes its confusion matrix and spamming behavior—could motivate a continuous-population inference problem. It would be a new, plausible question, but I cannot honestly present it as mirroring a named result in this paper or predict its place in the trichotomy on that basis.
That is the weak point, and here it is decisive: any prover-ready continuous question I supplied would lack the required anchor. The paper’s empirical accuracy and spammer-detection findings are not numbered complexity results.
There are no anchors to defeat: the paper states no numbered theorem, lemma, corollary, or proposition about the complexity of a computational problem. Its contributions are a Bayesian model, empirical accuracy and spammer-detection results, and reported running times on three finite datasets. None gives a complexity result for a population-continuous version. Under the requested standard, that is the strongest case against a *mirror of this paper’s results*.
The tempting improvement is to model a large annotation platform as a distribution over worker types, where a type captures a worker’s reliability or spamming behavior, and ask how efficiently one can infer the distribution of worker types and the documents’ category proportions. That is plausible: the paper already estimates worker confusion matrices and document proportions. But it would introduce a new population model and a new computational question. The paper’s categorical distributions \(\Lambda_i\) are distributions over document categories; they do not make the worker population continuous. The paper’s actual experiments instead use a finite set of workers and infer worker-specific confusion matrices.
There is a further tension in that proposed extension. If the aim includes identifying particular spammers for exclusion from future campaigns, the data attached to each worker matters. Grouping workers by latent behavior could still support an estimate of spammer mass, but it changes the individual-classification objective. That is not a decisive objection to a mass model; it does show that the attractive extension does not simply preserve every aim of MBCC.
So the negative case is strong against treating this paper as an anchored continuization candidate, but weak against the universal claim that no worthwhile continuous-worker problem could be built from its ideas. A large-crowd inference problem may be worth studying on its own. The honest conclusion is narrower: it would be a new project, not a continuous mirror of a computational result this paper established.
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.