{"id":"1607acd9-1db5-4f15-b5c4-5bfd4f78cd0e","arxiv_id":"2608.10247","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"DualSpectralCF attaches a signed user signal and a signed item-item operator to any spectral CF backbone, matching or beating its unsigned version on all five tested datasets with only two hyperparameters.","lead":"This paper adds two cheap components to spectral collaborative filtering that let it use explicit user dislikes without any training. The result is consistent accuracy gains, especially for cold-start users, at a small fraction of the cost of learned sign-aware models.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 'every instance matches or beats its unsigned backbone' claim is guaranteed by the evaluation protocol: the (0,0) cell in every (γ,κ) grid is the unsigned backbone, so choosing the best cell cannot lose; the fixed-default support covers only 4/5 datasets and the abstract mixes in an exception.","rationale":"Reader's weakest_assumption targets generalization of γ<0, but I think the selection protocol is more fundamental. Because (0,0) is in every grid, 'matches or beats' is a mathematical consequence of choosing the best cell. That does not make the paper worthless: the fixed-default ChebyCF results on four datasets, the full Table 6 sweeps, the honest cold-start breakdown in Table 10, and the BSPM bug fix are informative. However, the magnitude of the contribution — up to +32.6%, and even the +1.9% to +16.0% default range — depends on test-set selection; Table 5's per-dataset settings (KuaiRec γ=0, KuaiRand κ=1.0, etc.) were chosen after seeing test Recall@20. The paper would be conditionally acceptable if selection were moved to validation or if the default were evaluated on all five datasets without exceptions. I therefore retain the reader's CONDITIONAL verdict without change, with a sharper condition attached.","tokens_in":15699,"tokens_out":7668,"duration_ms":77209,"concrete_test":"Hold out test users before any hyperparameter choice. For each (dataset, backbone), tune backbone-native hyperparameters and (γ,κ) on a validation split only, then report test Recall@20 for the signed instance and for γ=κ=0 under the same backbone-native settings. Separately, force the fixed default (γ=−0.5, κ=0.1) on KuaiRec and report whether it beats the unsigned 0.0630. If fewer than 5/5 cells have positive validation-selected lifts, or if the forced-default KuaiRec result is below baseline, the 'every instance matches or beats' claim is an artifact of test-set selection.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The most load-bearing weakness is that the central claim — 'every instance matches or beats its unsigned backbone' — is guaranteed by the evaluation protocol rather than established empirically. In §3.1 the authors sweep γ ∈ {−0.5, −0.25, 0, 0.25} and κ ∈ {0, 0.1, 0.5, 1.0}, and Table 2 reports the best (γ,κ) per (dataset, backbone) cell. The point (0,0) is exactly the unsigned backbone and lies in every grid, so a cell-wise maximum can never be below baseline; the +3.2% to +32.6% lifts in Table 2 are maxima of a search, not evidence about a fixed configuration. The only non-tautological support is the 'fixed default' (γ=−0.5, κ=0.1) for ChebyCF, and that support is weaker than the abstract implies: Table 6(c) shows the default loses on KuaiRec (0.0417 vs unsigned 0.0630), so Table 5 instead uses γ=0 on KuaiRec, and the abstract's '+1.9% to +16.0% at the fixed default' includes this exception. The paper is transparent about the exception in §3.3 and Table 5, but the headline still rests on per-dataset test-set hyperparameter selection. A validation-based selection protocol, or a pre-registered default applied to all five datasets without exceptions, would be needed to know whether sign-awareness itself, rather than grid search, produces the reported lifts.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces DualSpectralCF, a training-free framework that extends spectral collaborative filtering backbones of the form r̂_u = F(M) r_u with two components: a signed input signal r_u± that encodes explicit negative feedback with weight −γ, and a signed item-item operator M± that blends like-together and dislike-together similarities with strength κ. The framework is instantiated on ChebyCF, GF-CF, and Turbo-CF and evaluated on five sign-aware benchmarks against unsigned spectral baselines, LightGCN, and SIGformer. The paper reports that every DualSpectralCF instance matches or beats its unsigned backbone, with per-cell best lifts up to +32.6%, and that DualSpectralCF-Cheby achieves +1.9% to +16.0% at the fixed default (γ=−0.5, κ=0.1) on four of five datasets. Additional results cover efficiency, cold-start users, a full 4×4 hyperparameter sweep, and a bug fix for the BSPM baseline.","tokens_in":16083,"tokens_out":4131,"duration_ms":43889,"significance":"If the empirical claims hold, the framework fills a genuine gap in the literature: it is the first sign-aware, training-free, backbone-agnostic spectral CF method, and the paper ships a clean abstraction, a public code repository, deterministic experiments, and a complete appendix with the full hyperparameter sweep and per-dataset numbers. The efficiency and Pareto-frontier arguments are plausible and well documented. However, the central quantitative claim—that every instance matches or beats its unsigned backbone—is currently an artefact of the evaluation protocol rather than a demonstrated property of a fixed configuration, because the hyperparameter grid always contains the unsigned backbone at (γ=0, κ=0). The paper is transparent about the KuaiRec exception, but the abstract and takeaways do not consistently carry that qualification. The significance of the method is therefore real but conditional on a re-analysis that separates selection from evaluation.","major_comments":[{"comment":"The claim that \"every DualSpectralCF instance matches or beats its unsigned backbone\" is guaranteed by the evaluation protocol rather than established empirically. In §3.1 the authors sweep γ ∈ {−0.5, −0.25, 0, 0.25} and κ ∈ {0, 0.1, 0.5, 1.0}, and Table 2 reports the best (γ,κ) per (dataset, backbone) cell. The point (0,0) is exactly the unsigned backbone and is contained in every grid, so the cell-wise maximum can never be below baseline; the +3.2% to +32.6% lifts are maxima of a search, not measurements at a fixed configuration. To support the abstract's claim, the paper should report lifts at a pre-specified default applied uniformly, or at hyperparameters selected on a validation split, with the test-set numbers reported separately.","section":"§3.1, Table 2, Table 6"},{"comment":"The fixed-default support is weaker than the abstract implies. The abstract states \"+1.9% to +16.0% for DualSpectralCF-Cheby at the fixed default (γ=−0.5, κ=0.1)\" without noting that this holds on only 4 of 5 datasets. On KuaiRec, Table 6(c) shows the default gives Recall@20 of 0.0417 versus 0.0630 for the unsigned backbone (γ=0, κ=0), so the default loses on that dataset; Table 5 consequently uses γ=0 on KuaiRec. The paper is transparent about this in §3.3 and Table 5, but the headline claim, the abstract, and the RQ2 takeaway should all carry the same qualification, or the authors should adopt a genuinely fixed default across all five datasets.","section":"Abstract, §3.3, Table 6(c), Table 5"},{"comment":"The \"why γ<0 works\" explanation is a post-hoc interpretation rather than a tested prediction. The finding that the best γ is non-positive in all 15 cells is obtained from the same data used to select the hyperparameters, and the KuaiRec cells for the Laplacian backbones prefer γ=0, meaning the negative channel is sometimes ignored entirely. The \"topical attention\" account in §2.2 is plausible, but it is not independently validated; as stated, it would predict that γ<0 helps across datasets. To make this a general property rather than an observed pattern, the authors should evaluate the default configuration on held-out data or additional sign-aware datasets, or explicitly label the explanation as one interpretation of the observed grid-search pattern.","section":"§2.2, §3.3, Table 6"}],"minor_comments":[{"comment":"The phrase \"every instance matches or beats its unsigned backbone on all 5 datasets\" should be replaced with a statement that clearly says this holds at per-cell selected hyperparameters, and the fixed-default claim should be qualified with \"on 4 of 5 datasets.\"","section":"Abstract and §3.2"},{"comment":"The term \"topical attention\" is introduced as an explanatory metaphor without a formal definition or a direct measurement; consider relabeling it as an interpretation or providing user-level evidence that negative items are topically similar to positive items.","section":"§2.2"},{"comment":"The caption should state explicitly that the reported lift is the best over the 4×4 sweep and that the corresponding (γ,κ) settings differ across cells; Table 6 shows this, but Table 2 alone is easy to misread as a single configuration.","section":"Table 2"},{"comment":"The BSPM bug-fix description is clear and valuable, but the phrase \"latent bug\" should be softened to \"implementation issue\" because the behavior depends on the preprocessing convention for zero-degree items, which the original paper's datasets do not exhibit.","section":"Appendix I"}],"recommendation":"major_revision","confidential_remarks":"The paper is well within the scope of CIKM and the framework is elegant and well documented. The stress-test concern is valid: the headline 'matches or beats' claim is an artefact of choosing the best cell from a grid that contains the baseline. I recommend a major revision that (1) reports results at a fixed default configuration or at validation-selected hyperparameters, (2) carries the KuaiRec exception consistently through the abstract and takeaways, and (3) either validates the γ<0 generalization on held-out data or presents it explicitly as a post-hoc observation. With those changes, the paper's contribution would be solid."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a clean, useful idea with an honest write-up, but the central empirical claim is weaker than the abstract makes it sound.\n\nThe two-component construction — a signed input row r±_u and a signed item-item operator M± — plugs into any spectral CF backbone of the form r̂_u = F(M) r_u. The genuinely new observation is empirical: across all 15 (dataset, backbone) cells, the best γ is non-positive, meaning explicit dislikes behave like small positive signals rather than suppression. That is a finding worth replicating. The experiments are largely transparent: full 4×4 sweeps in the appendix, deterministic baselines, a documented BSPM bug fix, and a code link. Credit where due.\n\nThe soft spot is the headline, \"every instance matches or beats its unsigned backbone.\" It is guaranteed by the protocol. The (0,0) cell in each sweep is exactly the unsigned backbone, and the paper reports the best cell in each 4×4 grid. Choosing the max of a search that contains the baseline cannot lose. The only non-tautological support is the fixed default for ChebyCF, but that default is applied to 4 of 5 datasets; KuaiRec switches to γ=0, and the abstract's \"+1.9% to +16.0% at the fixed default\" hides this exception. I don't think the authors intend to deceive — §3.3 and Table 5 disclose the exception — but the abstract overstates the support. A validation-based selection protocol, or a genuinely pre-registered default used on all five datasets, would settle whether sign-awareness itself helps. There is also a minor internal inconsistency: some unsigned baseline Recall@20 values in Table 1 differ from the γ=κ=0 cells in Table 6 (e.g., GF-CF and Turbo-CF on Amazon-CDs). The paper should explain that.\n\nThe cold-start section is cherry-picked but disclosed: Table 10 shows the lift turns negative for active users, and §3.5 says so. That is acceptable if the claim is scoped to cold-start.\n\nBottom line: the framework is simple, cheap, and plausible; the negative-γ finding is worth a serious look. But the accuracy claim needs validation-based selection or a pre-registered default before it can be taken at face value. This deserves peer review — a referee should ask for a fixed configuration evaluated on all five datasets and an abstract that matches what was actually run.","headline":"Clean, backbone-agnostic way to add explicit dislikes to training-free spectral CF, with a genuinely new negative-γ finding, but the headline accuracy claim is partly guaranteed by grid search over the unsigned baseline.","tokens_in":16550,"tokens_out":3117,"would_cite":true,"duration_ms":28701,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper claims that explicit negative feedback can be folded into any closed-form spectral collaborative-filtering backbone, without gradient training, and that doing so matches or beats the sign-blind backbone on every dataset tested.","keywords":["collaborative filtering","recommender systems","spectral graph filters","training-free methods","negative feedback","sign-aware recommendation","cold-start users"],"falsifier":"Construct a dataset with the same interaction counts but with dislike edges randomly permuted across items, or measure the subspace overlap between the positive and negative co-interaction gram matrices on an existing benchmark; the paper's attention mechanism predicts the optimal $\\gamma$ stays non-positive and the signed model beats the unsigned backbone, whereas the random-label version should drive the optimal $\\gamma$ to zero or positive and erase the gain.","tokens_in":15519,"feed_emoji":"👍","tokens_out":8497,"duration_ms":77771,"temperature":0.7,"pith_summary":"This paper claims that explicit negative feedback can be added to any closed-form spectral collaborative-filtering backbone by changing two ingredients of the scoring rule $\\hat{\\mathbf{r}}_u = F(\\mathbf{M})\\mathbf{r}_u$: the user's input row becomes a signed vector $\\mathbf{r}_u^{\\pm}$ that carries a weight $-\\gamma$ for items the user disliked, and the item-item operator becomes a signed version $\\mathbf{M}^{\\pm}$ that adds $\\kappa$ times the dislike-together similarity. No gradients are needed, only two scalar hyperparameters. Across five sign-aware benchmarks and three different spectral backbones, the paper reports that every instance matches or beats its unsigned backbone, with Recall@20 gains up to +32.6% under per-backbone tuning and +1.9% to +16.0% for the Chebyshev instance at a fixed default. The same instances run 7.7 to 155.3 times faster than a learned sign-aware transformer while reaching 70.7% to 90.7% of its accuracy, and the largest gains appear for cold-start users. A reader should care because this turns a signal that previously required expensive learned models into a cheap, closed-form upgrade to an existing family of recommenders.","feed_headline":"Dislikes boost spectral recommenders with zero training","feed_subtitle":"Adding two scalar weights to closed-form graph filters beats their sign-blind backbones on five benchmarks.","key_machinery":"The load-bearing identity is the spectral scoring template $\\hat{\\mathbf{r}}_u = F(\\mathbf{M})\\mathbf{r}_u$, where $\\mathbf{r}_u$ is the user's binary interaction row, $\\mathbf{M}$ is an item-item operator (either a normalized similarity $\\hat{\\mathbf{P}}=\\tilde{\\mathbf{R}}^{\\top}\\tilde{\\mathbf{R}}$ or a Laplacian $\\mathbf{L}_*=\\mathbf{I}-\\tilde{\\mathbf{R}}^{\\top}\\tilde{\\mathbf{R}}$), and $F$ is a scalar-valued graph filter. Component A replaces the input row by $\\mathbf{r}_u^{\\pm}$, and Component B replaces the operator by the signed Laplacian $\\mathbf{L}^{\\pm} = \\mathbf{I} - \\tilde{\\mathbf{R}}_+^{\\top}\\tilde{\\mathbf{R}}_+ + \\kappa\\tilde{\\mathbf{R}}_-^{\\top}\\tilde{\\mathbf{R}}_-$ for Laplacian backbones, or by the signed similarity $\\hat{\\mathbf{P}}^{\\pm} = \\tilde{\\mathbf{R}}_+^{\\top}\\tilde{\\mathbf{R}}_+ + \\kappa\\tilde{\\mathbf{R}}_-^{\\top}\\tilde{\\mathbf{R}}_-$ for similarity backbones. Each component modifies a different slot of the same template, which makes the pair independent and composable, and because the signed operator keeps the unsigned operator's sparsity pattern and polynomial-filter backbones need no eigendecomposition, the asymptotic cost remains $O(K \\cdot \\mathrm{nnz}(\\mathbf{R}))$ for a degree-$K$ filter.","core_discovery":"The paper's central claim is that the shared prediction template $\\hat{\\mathbf{r}}_u = F(\\mathbf{M})\\mathbf{r}_u$ used by training-free spectral collaborative-filtering methods can be made sign-aware by two substitutions: a signed input row $\\mathbf{r}_u^{\\pm}$ with entries $+1$ for likes, $-\\gamma$ for explicit dislikes, and $0$ otherwise; and a signed item-item operator $\\mathbf{M}^{\\pm}$ that blends the like-together structure $\\tilde{\\mathbf{R}}_+^{\\top}\\tilde{\\mathbf{R}}_+$ with a scaled dislike-together term $\\kappa\\tilde{\\mathbf{R}}_-^{\\top}\\tilde{\\mathbf{R}}_-$. The paper claims these substitutions are independent, backbone-agnostic, add only two hyperparameters $(\\gamma,\\kappa)$, and preserve each backbone's asymptotic cost. Empirically, every instantiation on three heterogeneous spectral backbones matches or beats its unsigned backbone on all five sign-aware benchmarks; at backbone-specific tuning the Recall@20 lift reaches +32.6%, and at a fixed default $(\\gamma=-0.5,\\kappa=0.1)$ the Chebyshev-based instance improves by +1.9% to +16.0%. The paper further claims that the optimal $\\gamma$ is never positive across all 15 dataset-backbone cells, interpreting this as evidence that an explicit dislike reveals topical attention even when the experience was bad, and that sign-awareness helps most for cold-start users, with up to +29.2% Recall@20 on Epinions users who have only 1 to 5 training items.","pith_inferences":["If the \"dislike reveals topical attention\" mechanism is general, the same signed-input substitution should transfer to other closed-form or linear recommenders that score items by filtering a user profile, not just spectral CF backbones, making sign-awareness a general property of linear scoring models.","The paper's KuaiRec result implies a testable monotonicity: the optimal $\\gamma$ should approach zero as negative labels get noisier or less topically informative, and an automated estimator of negative-signal quality (for example, from rater disagreement or label entropy) could set $\\gamma$ per dataset, which the paper leaves as future work.","A direct spectral test of the mechanism would compare the principal eigenspaces of $\\tilde{\\mathbf{R}}_+^{\\top}\\tilde{\\mathbf{R}}_+$ and $\\tilde{\\mathbf{R}}_-^{\\top}\\tilde{\\mathbf{R}}_-$: if dislikes mark attention, the negative gram matrix should share a substantial subspace with the positive one, and this overlap should predict when Component A helps.","Because Component A and Component B are independent, one can extend the framework to asymmetric sign encodings, such as a negative $\\kappa$ that amplifies dislike-together similarity rather than penalizing it, to test whether co-dislikes ever capture useful structure on their own."],"forward_implications":["Any future spectral CF backbone expressible as $\\hat{\\mathbf{r}}_u = F(\\mathbf{M})\\mathbf{r}_u$ can be made sign-aware with the same two substitutions, inheriting the reported gains without retraining.","Explicit negative feedback no longer requires a learned sign-aware model: a closed-form spectral filter with two scalars can approach a learned transformer's accuracy at 7.7 to 155.3 times less wall-clock time.","The strongest payoff is for cold-start users with 1 to 5 positive interactions, where Recall@20 rose up to +29.2%, suggesting negative history can partially substitute for sparse positive history.","A fixed global weight on dislikes is not the final answer: the paper observes the lift can become negative for very active users (up to $-11.1\\%$ on KuaiRec), which directly motivates per-user adaptive weighting.","The default $(\\gamma=-0.5,\\kappa=0.1)$ transfers across datasets for the Chebyshev-based instance on 4 of 5 benchmarks, indicating the recipe is robust across feedback types from star ratings to hate-button clicks."],"supporting_citations":[{"why":"Supplies the five sign-aware benchmarks, the learned sign-aware transformer baseline that the method is compared against, and the full-ranking evaluation protocol.","marker":"[2]"},{"why":"Defines the spectral CF formulation and one of the three backbones; the shared $\\hat{\\mathbf{r}}_u = F(\\mathbf{M})\\mathbf{r}_u$ template comes from this line.","marker":"[17]"},{"why":"Provides the Chebyshev-interpolated filter backbone on which DualSpectralCF-Cheby is instantiated.","marker":"[9]"},{"why":"Provides the polynomial-filter backbone on which DualSpectralCF-Turbo is instantiated.","marker":"[13]"},{"why":"Source of the Amazon-CDs and Amazon-Music datasets whose 1-star reviews define the negative edges.","marker":"[12]"},{"why":"Source of the Epinions dataset with user-to-user distrust serving as explicit negative feedback.","marker":"[20]"},{"why":"Source of KuaiRec, whose very-low watch-ratio negatives define the benchmark where sign-awareness is weakest.","marker":"[5]"},{"why":"Source of KuaiRand, whose hate-button clicks define another sign-aware benchmark used in the evaluation.","marker":"[6]"},{"why":"Establishes the signed Laplacian eigenvalue bound $[0,2]$ that keeps the low-pass filter well-posed when $\\kappa \\ge 0$.","marker":"[18]"}],"fun_headline_variants":["Two scalars make spectral CF sign-aware, training-free","Dislike-aware spectral filters beat sign-blind backbones","Training-free CF gains from explicit dislikes","Cold-start users gain most from dislikes in spectral CF","Spectral CF learns dislikes with zero training"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that explicit dislikes are topically informative: a user who hates a horror movie still reveals an interest in horror, so a small positive weight on disliked items helps; if dislikes are random or purely aversive, Component A reduces accuracy, which is exactly what the paper reports on KuaiRec, where the Laplacian backbones choose $\\gamma=0$ and fall back to the unsigned model.","fun_headline_variants_meta":{"raw":{"variants":["Two scalars make spectral CF sign-aware, training-free","Dislike-aware spectral filters beat sign-blind backbones","Training-free CF gains from explicit dislikes","Cold-start users gain most from dislikes in spectral CF","Spectral CF learns dislikes with zero training"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000503,"raw_usage":{"total_tokens":2595,"prompt_tokens":1218,"completion_tokens":1377,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":834,"completion_tokens_details":{"reasoning_tokens":1302}},"tokens_in":834,"tokens_out":1377,"duration_ms":10571,"temperature":1.0,"reasoning_tokens":1302,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T04:10:51.402351+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Construct a dataset with the same interaction counts but with dislike edges randomly permuted across items, or measure the subspace overlap between the positive and negative co-interaction gram matrices on an existing benchmark; the paper's attention mechanism predicts the optimal $\\gamma$ stays non-positive and the signed model beats the unsigned backbone, whereas the random-label version should drive the optimal $\\gamma$ to zero or positive and erase the gain.","supporting_citations":[{"cited_title":"How Powerful is Graph Convolution for Recommendation?","cited_arxiv_id":"2108.07567","evidence_quote":"Defines the spectral CF formulation and one of the three backbones; the shared $\\hat{\\mathbf{r}}_u = F(\\mathbf{M})\\mathbf{r}_u$ template comes from this line."},{"cited_title":"Turbo-CF: Matrix Decomposition-Free Graph Filtering for Fast Recommendation","cited_arxiv_id":"2404.14243","evidence_quote":"Provides the polynomial-filter backbone on which DualSpectralCF-Turbo is instantiated."},{"cited_title":"KuaiRec: A Fully-observed Dataset and Insights for Evaluating Recommender Systems","cited_arxiv_id":"2202.10842","evidence_quote":"Source of KuaiRec, whose very-low watch-ratio negatives define the benchmark where sign-awareness is weakest."},{"cited_title":"KuaiRand: An Unbiased Sequential Recommendation Dataset with Randomly Exposed Videos","cited_arxiv_id":"2208.08696","evidence_quote":"Source of KuaiRand, whose hate-button clicks define another sign-aware benchmark used in the evaluation."}],"review_version":1}