{"id":"eeae61ca-e448-4c19-8f42-94907069185f","arxiv_id":"2607.02872","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"low","formal_verification":"none","parameter_count":4,"one_line_summary":"A Gamma-Markov multi-relational Poisson model with an explicit relation-transition kernel improves sparse dynamic KG link prediction and recovers interpretable inter-relation dynamics.","lead":"PGRE is a Bayesian Poisson-Gamma model that jointly tracks entity communities and how relation types transition over time in dynamic knowledge graphs. It improves sparse link prediction and yields interpretable relation-evolution patterns on standard TKG benchmarks.","discovery_kind":"extension","skeptic_critique":{"model":"grok-4.5","headline":"Superior AUC-PR claims rest on heavily filtered 500-entity/top-relation subgraphs whose inter-relational structure may not match the original TKGs.","rationale":"The reader correctly isolates the aggressive entity/relation filtering of §5.1 as the weakest link supporting the strongest claim. All other components—the Poisson–Bernoulli construction, the conjugate CRT-augmented Gibbs sampler, the explicit Π kernel, and the reported sparse-regime gains—are internally consistent and well-supported by the ablations and diagnostics. No deeper mathematical inconsistency or unacknowledged modeling error appears. Because the concern is already the basis of the CONDITIONAL verdict, no adjustment is required; the concrete test above simply makes the sensitivity check explicit and falsifiable.","tokens_in":17787,"tokens_out":586,"duration_ms":19458,"concrete_test":"Re-run the ICEWS18 pipeline of §5.1 with the same top-25 relations but the top-2000 (instead of 500) most active entities, retrain PGRE and the two strongest baselines (D-NGPPF, RENet) under identical Gibbs/epoch budgets, and recompute Table-1-style AUC-PR/AUC-ROC at t=T. If PGRE’s absolute AUC-PR drops by >30 % relative or its margin over the baselines shrinks below 0.02, the filtering assumption is load-bearing and the headline claim weakens.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim (competitive/superior link prediction, especially AUC-PR, on sparse multi-relational TKGs via the relation-transition kernel inside the Gamma-Markov process) is demonstrated exclusively on the moderate-scale tensors of §5.1: top-frequency relations (R=8–15) followed by the 500 most active entities. This selection can artificially densify local neighborhoods, induce spurious cross-relation co-occurrences among high-degree entities, and alter the very transition statistics that Π is meant to capture. Consequently the large AUC-PR lifts in Table 1 (ICEWS18 0.198 vs. next-best 0.127; GDELT 0.277 vs. 0.136) and the interpretability patterns in Figs. 4–5 may not survive on the unfiltered graphs or on standard full-scale TKG protocols that use MRR/Hits@k with negative sampling. The R-ablation of Fig. 3 keeps the entity set fixed, so it does not test this assumption. Without evidence that the filtered subgraphs preserve the target inter-relational dynamics, the claimed advantage on “sparse multi-relational temporal knowledge graphs” remains conditional on a preprocessing regime that is not representative of the original data.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.5","summary":"The paper proposes PGRE, a Bayesian generative model for dynamic multi-relational knowledge graphs. Multi-relational temporal links are generated via a Bernoulli–Poisson (equivalently Poisson–Bernoulli) construction whose intensity factors into subject/object Gamma entity–community loadings and time- and relation-specific community weights. Those weights evolve under a Gamma–Markov process whose shape is a linear mixture of previous-relation weights through a Dirichlet-prior transition matrix Π, so that inter-relational temporal dependence is modeled explicitly inside shared latent communities. Posterior inference uses Gibbs sampling with negative-binomial–logarithmic / CRT data augmentation and standard Poisson–multinomial and Gamma–Poisson conjugacies (Theorems 1–3, §4 and Supp. §7.5). On moderate-scale filtered versions of ICEWS18, GDELT, and WIKI, PGRE reports competitive or superior AUC-PR (especially on the sparser multi-relational sets), an ablation removing Π (PGE), a relation-cardinality study, runtime comparisons, and visualizations of latent-factor trajectories and transition structure.","tokens_in":18133,"tokens_out":1739,"duration_ms":22665,"significance":"Bayesian modeling of dynamic multi-relational knowledge graphs with an explicit relation-transition kernel is relatively underexplored compared with neural TKG models and single-relation Poisson–Gamma network models. If the empirical advantage holds under more representative evaluation, PGRE would be a useful, interpretable alternative for sparse multi-relational temporal settings, with closed-form conjugacy-based Gibbs updates, public code, an ablation isolating Π, τ-sensitivity checks, and sampling diagnostics. The technical extension from single-relation Gamma–Markov community models to a multi-relational transition kernel is a clear, incremental contribution to the probabilistic-network literature.","major_comments":[{"comment":"§5.1 preprocessing and Table 1: All reported gains (notably AUC-PR on ICEWS18 0.198 vs. next-best 0.127 and GDELT 0.277 vs. 0.136) are obtained after retaining only the most frequent relations (R=8–15) and then the 500 most active entities. This selection can densify local neighborhoods and reshape the very cross-relation co-occurrence and transition statistics that Π is intended to capture. The R-ablation in Fig. 3 keeps the entity set fixed, so it does not test this assumption. The central claim of advantage on “sparse multi-relational temporal knowledge graphs” therefore remains conditional on a non-standard filtered regime. Please either (i) report results on unfiltered or substantially larger entity sets under the same protocol, or (ii) provide quantitative evidence that the filtered tensors preserve the inter-relational temporal structure of the original TKGs, and state the limitat","section":"§5.1, Table 1, Fig. 3"},{"comment":"§5.1 evaluation protocol: Performance is reported only as AUC-ROC/AUC-PR on a single future snapshot (tq=T) with binary presence labels. Standard TKG forecasting literature typically reports filtered MRR and Hits@k under negative sampling (or time-aware ranking). Without at least one ranking-based metric on the same splits—or a clear argument why AUC is the right primary metric for this generative formulation—it is hard to place the “competitive or superior” claim relative to the neural baselines the paper cites (Know-Evolve, DyRep, RE-NET). Adding MRR/Hits@k (or an equivalent ranking evaluation) on the same data would make the comparison load-bearing rather than protocol-dependent.","section":"§5.1 Experimental Settings, Table 1"},{"comment":"Supp. §7.1 and Table 1 baselines: Single-relation probabilistic models (DPGM, D-NGPPF, G-HSEPM) are trained independently per relation on bipartite undirected conversions, while PGRE jointly models all relations through Π and shared communities. That design difference is legitimate but should be discussed as a confound when attributing AUC-PR gains specifically to the transition kernel rather than to joint multi-relational sharing alone. The PGE ablation helps, but a joint multi-relational baseline without temporal Π (or with a diagonal-only Π) on the same tensor would isolate the contribution more cleanly.","section":"Supp. §7.1, Table 1, §5.2 PGE ablation"},{"comment":"Supp. §7.3 sampling diagnostics and Figs. 4–5: Post-burn-in ESS for some normalized community weights is very low (e.g., 12.99 and 23.63), with strong serial dependence in the trace plots. Interpretability claims that rest on temporal evolution of δ^{(t,r)}_k and on the structure of Π (Figs. 4–5, §5.2) need either longer chains / thinning, multiple-chain diagnostics, or a statement that those figures are qualitative and not fully mixed posterior summaries. Low ESS does not invalidate link-prediction means if predictive functionals mix better, but it weakens the “revealing meaningful relational evolution patterns” claim as currently stated.","section":"Supp. §7.3, Figs. 4–5, §5.2"}],"minor_comments":[{"comment":"Abstract and §1 claim “particularly in sparse settings,” but §5.1 deliberately keeps the most active entities; clarify what “sparse” refers to (relation density, positive rate, or entity activity) and align wording with the constructed tensors.","section":"Abstract, §1, §5.1"},{"comment":"Eq. (9) Dirichlet prior uses a special diagonal mass ξν_r; the role of ξ versus the off-diagonal ν_r'ν_r construction is easy to miss—add a short sentence on why self-transitions are parameterized separately.","section":"§3.2, Eq. (9)"},{"comment":"Fig. 1 caption uses y_uv,t while the model uses M^{(t,r)}_{ij} / x^{(t,r)}_{ij}; unify notation between the motivational figure and the generative model.","section":"Fig. 1, §3.2"},{"comment":"Table 2: PGRE is fastest on WIKI but slower than several count models on GDELT; a one-sentence complexity discussion (dependence on R, K, T, and nonzero events) would help readers interpret the runtime trade-off.","section":"Table 2, §5.1"},{"comment":"Supp. Algorithm 1 and the main-text sampling order for ρ^{(t)}_k / forward CRT variables are slightly hard to follow; a short forward–backward sketch in the main §4 would improve readability without repeating the full supplement.","section":"§4, Supp. Algorithm 1"},{"comment":"Related work could briefly position against other multi-relational Bayesian TKG / tensor models beyond the count-reshaping line (PGDS/PRGDS/NBRGDS), even if only to note differences in how relation interactions are encoded.","section":"§2"}],"recommendation":"major_revision","confidential_remarks":"The technical core (Gamma–Markov multi-relational extension + conjugacy-based Gibbs) looks sound and is a natural next step from the authors’ prior single-relation Poisson–Gamma line. The main risk for the journal is overstated generality: results live on heavily filtered 500-entity tensors with AUC-only evaluation. If the authors can add a larger-scale or ranking-based experiment and tone the abstract claims accordingly, this is a solid methods paper; if they cannot, the contribution is still real but narrower than the title/abstract suggest. Fit is appropriate for a probabilistic ML / network-modeling venue; borderline for a pure TKG venue that expects MRR/Hits@k on full benchmarks."},"author_rebuttal":null,"desk_editor":{"model":"grok-4.5","letter":"The one thing worth knowing: this is a competent Bayesian extension of the Poisson–Gamma / Gamma–Markov line (D-NGPPF, DPGM, PGDS family) to multi-relational temporal knowledge graphs. What is actually new is the Dirichlet-governed relation-transition matrix that couples community weights across relation types inside the same Gamma process, plus the conjugate Gibbs machinery that makes it runnable.\n\nThey do the technical work carefully. The conjugacy arguments and data-augmentation steps are standard and correctly applied; the sampler is written out in full in the supplement; there is a real ablation (PGE without the transition matrix) that hurts AUC-PR; τ sensitivity is stable; code is linked. On the filtered ICEWS18/GDELT/WIKI slices they post clear AUC-PR gains over both count-based probabilistic models and several neural TKG baselines, and the transition heatmaps plus latent-factor tracks are readable (strong self-persistence, a few semantic off-diagonals). That package is solid for the subfield.\n\nThe soft spot is real and matches the stress-test note, but it does not sink the paper. All main numbers come from top-frequency relations and the 500 most active entities. That selection can densify local structure and change the very transition statistics Π is meant to capture. The R-ablation keeps the entity set fixed, so it does not answer the filtering question. Evaluation is AUC-PR/ROC forecasting, not the usual MRR/Hits@k protocol, and K is fixed at 200. These keep the work in “solid extension” rather than “new standard” territory. Self-citations are to prior single-relation Poisson–Gamma work that is independently published; that is fine.\n\nWho it is for: people already in Bayesian dynamic networks or sparse TKG completion who want an explicit inter-relation kernel with closed-form updates. Not required reading for general graph ML.\n\nI would send it to peer review. The math is honest, the contribution is checkable, and the right referee push is full-scale or less-filtered experiments plus standard TKG metrics—not a desk reject. Engage if you work in this niche; otherwise skim the model section and the ablation.","headline":"Clean multi-relational Poisson–Gamma extension; the relation-transition kernel is the real addition, but the big AUC-PR lifts sit on heavily filtered 500-entity subgraphs.","tokens_in":18736,"tokens_out":561,"would_cite":true,"duration_ms":15843,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.5","headline":"A Poisson-Gamma model with an explicit relation-transition kernel predicts sparse temporal links and recovers how relations evolve.","keywords":["dynamic knowledge graphs","link prediction","Poisson-Gamma model","relation transition","Gamma-Markov process","Gibbs sampling","multi-relational temporal data"],"falsifier":"Re-run the identical forecasting protocol on the unfiltered full tensors (or on a random sample of low-frequency relations and low-activity entities) and check whether the AUC-PR gains of PGRE over the strongest baselines disappear or reverse.","tokens_in":18647,"feed_emoji":"🔗","tokens_out":791,"duration_ms":7963,"temperature":0.7,"pith_summary":"Dynamic knowledge graphs change over time and mix many relation types, yet most predictors treat those relations as independent and struggle when data are sparse. This paper proposes PGRE, a generative model that represents multi-relational temporal links with a Poisson-Bernoulli construction, attaches Gamma latent factors to entities as subjects and objects, and lets shared latent communities mediate cross-relation influence. A Gamma-Markov process equipped with a Dirichlet transition matrix then carries those community weights from one time step to the next, so that the probability of one relation can depend on the history of others. Gibbs sampling with negative-binomial and Chinese-restaurant-table augmentations yields closed-form updates. On filtered ICEWS18, GDELT and WIKI snapshots the model matches or exceeds strong neural and Bayesian baselines on AUC-PR, especially under sparsity, while the learned transition matrix and factor trajectories display self-persistence and structured switches among related relation types.","feed_headline":"Poisson-Gamma model tracks how relations evolve in sparse graphs","feed_subtitle":"An explicit transition kernel lifts link-prediction accuracy and shows which relations follow which.","key_machinery":"PGRE: a Poisson-Bernoulli generative model whose community weights evolve by a Gamma-Markov process whose shape is a linear mixture of all previous relation-specific weights, the mixing coefficients forming a Dirichlet-drawn transition matrix that couples relations across time.","core_discovery":"Jointly modeling entity-community factor loadings and an explicit relation-to-relation transition kernel inside a Gamma-Markov process produces competitive or superior link-prediction accuracy on sparse multi-relational temporal knowledge graphs and recovers interpretable patterns of relational evolution.","pith_inferences":["If the transition kernel can be made time-varying, the same framework could track regime shifts such as sudden changes in diplomatic or collaboration patterns.","The hierarchical Gamma shrinkage already present for communities suggests a natural route to nonparametric inference over the number of active relation types themselves.","Because inference is fully Bayesian and closed-form, the model could serve as a calibrated uncertainty source for hybrid systems that combine probabilistic forecasts with large language models."],"forward_implications":["Sparse multi-relational forecasting can be improved by an explicit, learnable relation-transition kernel rather than by deeper neural encoders alone.","Posterior samples of the transition matrix and community weights supply human-readable maps of how one relation type tends to follow another.","The same Gamma-Markov construction can be applied to other sparse multi-type count tensors that evolve over discrete time.","Ablation by removing the transition matrix measurably lowers AUC-PR, confirming that cross-relation coupling is a necessary ingredient of the reported gains."],"fun_headline_variants":["Poisson-Gamma model captures relation evolution in sparse dynamic graphs","Gamma-Markov process tracks inter-relational dependencies over time","PGRE links entity factors and relation transitions for sparse prediction","Explicit kernel reveals how multi-relational patterns evolve in DKGs","Joint Poisson-Bernoulli and Gamma factors boost temporal link accuracy"],"cache_read_input_tokens":128,"weakest_assumption_plain":"The filtered subgraphs that keep only the most frequent relations and the 500 most active entities still contain the same inter-relational temporal structure the model is meant to learn.","fun_headline_variants_meta":{"raw":{"variants":["Poisson-Gamma model captures relation evolution in sparse dynamic graphs","Gamma-Markov process tracks inter-relational dependencies over time","PGRE links entity factors and relation transitions for sparse prediction","Explicit kernel reveals how multi-relational patterns evolve in DKGs","Joint Poisson-Bernoulli and Gamma factors boost temporal link accuracy"]},"model":"grok-4.5","effort":"low","cost_usd":0.00294,"raw_usage":{"total_tokens":998,"prompt_tokens":666,"num_sources_used":0,"completion_tokens":84,"cost_in_usd_ticks":29400000,"prompt_tokens_details":{"text_tokens":666,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":248,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":666,"tokens_out":84,"duration_ms":3106,"temperature":1.0,"reasoning_tokens":248,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-12T06:27:38.658301+00:00","model_set":{"reader":"grok-4.5"},"falsifier":"Re-run the identical forecasting protocol on the unfiltered full tensors (or on a random sample of low-frequency relations and low-activity entities) and check whether the AUC-PR gains of PGRE over the strongest baselines disappear or reverse.","supporting_citations":[],"review_version":1}