Text Meets Topology: Rethinking Out-of-distribution Detection in Text-Rich Networks

Danny Wang; Guangdong Bai; Ruihong Qiu; Zi Huang

REVIEW 3 major objections 4 minor 1 cited by

Text Meets Topology: Rethinking Out-of-distribution Detection in Text-Rich Networks

T0 review · 3 major / 4 minor · reviewed 2026-08-15 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Fusing text and topology with cross-attention and node-specific projections catches out-of-distribution nodes in text-rich networks where post-hoc and graph-only baselines miss.

desk verdict TextTopoOOD is a genuinely useful benchmark, but the paper's 'consistently outperforms' claim is contradicted by its own per-scenario results on Reddit text shifts. read the letter →

arxiv 2508.17690 v2 pith:3G2RUGNO submitted 2025-08-25 cs.CL cs.LG

Danny Wang , Ruihong Qiu , Guangdong Bai , Zi Huang This is my paper · ORCID

classification cs.CLcs.LG

keywords out-of-distributiondetectiontext-richnetworksgraphneuralcross-attentionHyperNetworkenergy-basedscoringdistributionshiftbenchmark

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper sets out to show that out-of-distribution detection in text-rich networks is a two-modality problem: a node can drift because its language changed, because its connections changed, or because the two no longer agree. To test this it introduces TextTopoOOD, an evaluation suite covering attribute-level, structural, thematically-guided label, and domain-based shifts across 11 datasets. It then proposes TNT-OOD, which fuses neighborhood structure into text embeddings with cross-attention and projects each node through HyperNetwork-generated weights, scoring nodes by combining classifier energy with a text-topology alignment term. The reported results claim consistent gains over propagation-based and post-hoc baselines, with the largest improvements on citation and social networks.

What carries the argument

The load-bearing mechanism is the combination of neighborhood cross-attention fusion and node-specific HyperNetwork projection. Cross-attention uses structure embeddings as queries over neighboring text, injecting local graph context into each node's text representation; the HyperNetwork then emits per-node low-rank projection weights so that ID nodes can be aligned in a contrastive space while OOD nodes expose misalignment. The final detector score couples the energy of the classifier logits with an alignment score and propagates the result across the graph.

What would settle it

Compare TNT-OOD against its own no-propagation variant on a split where OOD nodes and all incident edges are withheld from the adjacency matrix at inference. If the AUROC and FPR95 gap over energy-alone shrinks to zero once OOD edges disappear, the reported gains depend on seeing the very connections that an inductive deployment would not have.

Watch

Extended reading notes

Core claim

The central claim is that TNT-OOD consistently outperforms or matches existing graph-OOD detectors across diverse network domains and OOD scenarios. The discovery is a design rather than a single identity: because OOD can surface as textual, structural, or joint text-structure misalignment, the detector builds a fused representation from a GCN-derived structure query and text keys and values via cross-attention, then generates node-specific projection matrices from a low-rank HyperNetwork, aligning text and structure embeddings with a symmetric contrastive loss. At test time the score subtracts a temperature-scaled alignment term from the energy logit score and propagates the smoothed scores over the adjacency matrix. The paper also argues that TextTopoOOD is the first benchmark to cover this range of shifts, and that the variance across scenarios shows the framework's difficulty.

Load-bearing premise

The detector assumes that at test time the full graph, including edges attached to out-of-distribution nodes, is already known and can be used to spread scores; if out-of-distribution nodes arrive without any recorded connections, the propagation step has nothing to operate on.

Editorial extensions

If this is right

In citation networks, TNT-OOD reduces the false-positive rate at 95% true-positive rate by up to roughly 10 percentage points over the compared baselines.
On knowledge and social networks such as Reddit and WikiCS, the method raises average AUROC markedly, for example from about 57.43 to 70.61 on Reddit.
Score propagation helps detection on Cora but hurts it on Arxiv, so the value of graph-based score spreading is dataset-dependent rather than universal.
Coupled text and structure shifts are detected more accurately than either shift alone, suggesting that real-world joint shifts may be easier to flag than isolated ones.
TextTopoOOD exposes genuinely hard cases: on Arxiv even TNT-OOD underperforms relative to its results elsewhere, marking the temporal and feature-shift scenarios as open challenges.
Beyond the paper: the full-graph inductive setting means the reported numbers assume OOD nodes' edges are visible at inference; a cold-start evaluation with disconnected OOD nodes would isolate whether the alignment score alone carries the signal.
Beyond the paper: because the alignment score measures agreement between projected text and structure-aware embeddings, replacing the frozen SBERT encoder with a different sentence encoder is a direct test of whether the gain depends on one embedding geometry.
Beyond the paper: node-specific projections from a HyperNetwork could serve as a learned calibration layer for any two-modality model, so the idea transfers to image-text or tabular networks even though the paper does not claim this.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

The benchmark is the real contribution here. TextTopoOOD covers four shift families (attribute, structural, thematically-guided label, domain) across 11 datasets, with code and construction details released. That is a step up from prior OOD-TAG/GLIP/GSyncOOD, which mostly stick to label or degree-based splits. If you work on graph or text-attributed OOD detection, this is a worthwhile evaluation suite to have on hand.

The method part is more ordinary. TNT-OOD is a reasonable composition of cross-attention, a HyperNetwork projection, and the energy-plus-propagation score from Wu et al. The ablation shows the cross-attention and contrastive loss help; the HyperNetwork is mixed. That is fine, but it does not change the fact that the central claim is over-stated. Section 5.1 says TNT-OOD 'consistently outperforms or matches baseline methods.' The paper's own Appendix I shows that on Reddit, Text (Synonym) and Text (Antonym), TNT-OOD gets AUROC 3.77 and 4.72, while GNNSafe gets 84.52 and NodeSafe gets 87.80/97.79. That is not a small underperformance; it is an inverted ranking. Table 2 hides it by aggregating across scenarios with a 37.62 standard deviation. The honest summary is that TNT-OOD is brittle on text-only shifts in at least one social network, not that it consistently beats baselines.

The full-graph-at-inference assumption is a second soft spot. Appendix D.1 says label-shift OOD nodes keep the complete edge set so the full graph is accessible during inference. The scoring function in Eq. 26-27 propagates over the adjacency matrix. If OOD nodes arrive without observed connections, the propagation step is undefined or silently becomes ID-only. The paper never flags this as a limitation, though the Limitations section does admit to approximation strategies and limited configuration search. Those admissions are honest but do not fix the main reporting problem.

Citation pattern is fine; the method credits Wu et al. for propagation and builds on prior graph OOD work. The LLM-as-detector comparison is methodologically muddy but clearly secondary.

Bottom line: worth a serious referee, but the authors need to revise the 'consistent' claim, report per-scenario results without hiding the Reddit failure, and discuss the inductive setting honestly. I would bring the benchmark to a reading group, and I would cite it for the benchmark, not for the method.

Referee Report

3 major / 4 minor

Summary. The paper introduces two contributions: TextTopoOOD, a benchmark framework for out-of-distribution detection in text-rich networks covering attribute-level text/feature shifts, structural shifts, thematically guided label shifts, and domain-based splits across 11 datasets; and TNT-OOD, a detection method that fuses textual and structural information through cross-attention and uses a HyperNetwork to generate node-specific projections, with an OOD score combining energy and text-topology alignment. The empirical study compares TNT-OOD against post-hoc and graph-specific baselines and reports aggregated AUROC/AUPR/FPR95 over the proposed scenarios.

Significance. If the benchmark is adopted by the community, TextTopoOOD would fill a real gap: existing graph OOD benchmarks focus on label or degree/temporal splits and largely ignore the coupled text-topology shifts that arise in social, e-commerce, and citation networks. The paper ships code, uses 11 datasets across four shift families, and reports three-seed means with ablations and hyperparameter sensitivity, which supports reproducibility. The method itself is reasonable and the alignment-score idea is clearly motivated. However, the central empirical claim of consistent superiority is contradicted by the paper's own per-scenario tables, and the test-time assumption of full graph access is not discussed as a limitation.

major comments (3)

[Section 5.1 and Appendix I, Table 20] The claim that TNT-OOD 'consistently outperforms or matches baseline methods' is not supported by the per-scenario results. On Reddit, for Text (Synonym), TNT-OOD obtains AUROC 3.77 ± 9.45 and FPR95 98.70 ± 28.31, while GNNSafe obtains 84.52 and NodeSafe obtains 87.80 AUROC; for Text (Antonym), TNT-OOD obtains 4.72 AUROC versus 88.52 and 97.79 for the two propagation baselines. An AUROC near 4 is worse than random and worse than every reported baseline, and it occurs on exactly the attribute-level text shifts that the paper's social-network motivation identifies as central. The aggregate Reddit row in Table 2 (70.61 ± 37.62) hides this failure because the standard deviation is enormous. The consistency claim should be removed or replaced with an explicit per-scenario discussion of where TNT-OOD fails and why.
[Section 4.2, Eq. (27), and Appendix D.1] The final OOD score is smoothed by K-layer propagation over the adjacency matrix, and Appendix D.1 states that 'the full graph structure remains accessible during inference with its associated OOD nodes.' This is a strong inductive assumption: if OOD nodes arrive without observed edges, or with only partially observed topology, Eq. (27) is undefined or degenerates to propagation over an ID-only graph. The manuscript does not list this as a limitation and does not evaluate a variant where OOD-node edges are withheld at test time. Since several real-world deployments of node-level OOD detection involve newly arriving nodes, this modeling choice should be explicitly acknowledged and ideally stress-tested.
[Section 5.1, Table 2] On Arxiv, TNT-OOD achieves AUROC 47.65, which is substantially below Energy (76.96) and NECO (75.19); the paper's statement that TextTopoOOD 'reveals challenging scenarios' explains the difficulty but does not address the fact that the proposed method is among the weakest on that dataset. The aggregated presentation in Table 2 should be supplemented with per-scenario counts of wins and losses against the best baseline, and the phrase 'significant improvements' should be limited to datasets and scenarios where the improvements actually hold.

minor comments (4)

[Table 2 caption] The caption says Best and Runner-up are highlighted, but the table does not visually distinguish the two; please specify the formatting used for each.
[Table 5] The row labels 'TNT-OODFPR@ 4o mini TPR(↓)' and 'TNT-OODFPR@ Gemini TPR(↓)' are garbled, and the metric description 'FPR at equivalent TPR levels' should be defined more precisely.
[Table 1 and Eq. (2)] There are typos: 'TEXTAUGMEN T' in Table 1's caption and 'a TnR' in the sentence below Eq. (2) should read 'a TrN'.
[Appendix D.1] For text augmentation the paper sets alpha_text = 1 and p_char = 1, meaning every eligible word is replaced and every replacement also receives character-level edits; this is a very strong perturbation, and it would help to report the actual coverage statistics for different datasets.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: TNT-OOD is not fitted to TextTopoOOD labels and its OOD score is not used to define the benchmark; the derivation chain is self-contained.

full rationale

Walking the derivation chain, TNT-OOD's OOD score (Eqs. 25-27) is a combination of classifier energy and a text-structure alignment term, and the model is trained only with ID classification and contrastive objectives (Eqs. 22-24). Hyperparameters are selected using ID classification performance, not OOD labels, so the detector is not statistically forced by the benchmark. The TextTopoOOD scenarios in Section 3 are generated from text augmentation, feature mixing, SBM rewiring, semantic-connection rewiring, text swaps, LLM-guided class selection, and temporal splits; none of these constructions uses TNT-OOD's scores, learned projections, or alignment term. The semantic-connection shift does consume the same frozen SBERT embeddings that TNT-OOD also uses, but that is a shared input representation rather than an equation-level reduction of the OOD prediction to the benchmark label. Self-citations (GOLD, CAT, PUMA) appear only in related work and are not load-bearing; no uniqueness theorem or ansatz is imported from the authors' prior work. The paper's own per-scenario results (e.g., Table 20, Reddit Text Synonym AUROC 3.77) contradict the abstract's and Section 5.1's aggregated claim of consistent superiority, and the full-graph inductive assumption in Appendix D.1 is a genuine limitation, but those are correctness and robustness concerns, not circularity. The benchmark and method are co-designed, which limits external validation, but the derivation itself is not equivalent to its inputs, so the appropriate circularity finding is none.

Assumptions & free parameters 7 free parameters · 5 assumptions · 0 invented entities

The method is a learned neural model, so the parameter ledger records the hand-chosen and grid-searched hyperparameters that shape the detector and the benchmark. The axioms capture the assumptions that the energy score, score propagation, synthetic shifts, full-graph test-time access, and frozen text embeddings are all valid for the TrN OOD problem. These are not derived in the paper, and the full-graph access assumption is the most fragile.

free parameters (7)

lambda (contrastive loss weight) = 0.5 for Cora/PubMed/WikiCS area (Table 11), 0.1 or 1 for others, selected via grid search
Controls the weighting of the contrastive alignment loss vs classification loss; affects final representations and OOD scores. Not derived from theory.
tau (contrastive temperature) = 0.1 selected for Bookhis and Elecomp (Table 12)
Temperature in the symmetric contrastive loss (Eq. 22); tuned per dataset on ID accuracy.
Score propagation steps K and alpha_score = K=3, alpha_score=0.5 for all methods
Number of propagation iterations and concentration parameter for the final score smoothing (Eq. 27), inherited from GNNSafe; set heuristically.
Text augmentation severity (alpha_text, p_char) = alpha_text=1, p_char=1
In Appendix D.1 the authors set these to 1 for all datasets to create 'sufficient diversity'. This makes the attribute-level shift very aggressive and is a hand-chosen parameter of the benchmark.
SBM rewiring parameters (beta, f_ii, f_ij) = Mild: beta=0.2, fii=0.7, fij=0.5; Medium: beta=0.5, fii=0.6, fij=0.3; Strong: beta=1.0, fii=0.4, fij=0.7
Control the structural OOD intensity in the benchmark (Eq. 6-7); chosen by hand to define difficulty levels.
Semantic connection similarity thresholds = 0.75, 0.85, 0.95
Thresholds for selecting edges in the semantic connection shift (Eq. 8); define how similar/dissimilar connected nodes are in OOD graphs.
Text swap ratio beta_swap = 1.0
Proportion of nodes whose text is swapped in the text swap shift (Eq. 9); set to maximum to create maximal text-structure mismatch.

assumptions (5)

domain assumption Energy score (Eq. 2) is a valid OOD scoring function for node-level detection in text-rich networks.
Adopted from Liu et al. 2020 without re-validation for TrNs; the paper assumes it transfers.
domain assumption Score propagation (Eq. 3, Eq. 27) improves OOD detection by smoothing scores over the graph.
Inherited from GNNSafe/NODESafe; the paper uses K=3, alpha=0.5 and does not test whether this holds when OOD nodes are isolated.
ad hoc to paper The TextTopoOOD synthetic shifts are representative of real-world OOD in text-rich networks.
The benchmark is constructed by the authors with hand-picked intensity parameters and LLM-chosen class splits; no external validation that these match real-world OOD.
domain assumption The full graph, including edges incident to OOD nodes, is available at inference.
Stated in Appendix D.1: OOD data maintains the complete edge set so the full graph structure is accessible during inference. This is essential for propagation-based scoring.
domain assumption Frozen SBERT embeddings (All-MiniLM-L6-v2) capture sufficient textual semantics for OOD detection.
The paper keeps the text encoder fixed and only varies the graph encoder; it assumes the fixed embedding space is adequate for detecting textual shifts.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Text Meets Topology: Rethinking Out-of-distribution Detection in Text-Rich Networks." pith.science (2026). https://pith.science/paper/3G2RUGNO

@misc{pith2026250817690,
  author       = {Pith},
  title        = {Pith review of: Text Meets Topology: Rethinking Out-of-distribution Detection in Text-Rich Networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3G2RUGNO}},
  note         = {Machine review of arXiv:2508.17690}
}

read the original abstract

Out-of-distribution (OOD) detection remains challenging in text-rich networks, where textual features intertwine with topological structures. Existing methods primarily address label shifts or rudimentary domain-based splits, overlooking the intricate textual-structural diversity. For example, in social networks, where users represent nodes with textual features (name, bio) while edges indicate friendship status, OOD may stem from the distinct language patterns between bot and normal users. To address this gap, we introduce the TextTopoOOD framework for evaluating detection across diverse OOD scenarios: (1) attribute-level shifts via text augmentations and embedding perturbations; (2) structural shifts through edge rewiring and semantic connections; (3) thematically-guided label shifts; and (4) domain-based divisions. Furthermore, we propose TNT-OOD to model the complex interplay between Text aNd Topology using: 1) a novel cross-attention module to fuse local structure into node-level text representations, and 2) a HyperNetwork to generate node-specific transformation parameters. This aligns topological and semantic features of ID nodes, enhancing ID/OOD distinction across structural and textual shifts. Experiments on 11 datasets across four OOD scenarios demonstrate the nuanced challenge of TextTopoOOD for evaluating OOD detection in text-rich networks.

Figures

Figures reproduced from arXiv: 2508.17690 by the authors.

**Figure 1.** Overview of TextTopoOOD framework. networks, nodes represent products with textual descriptions, connected by co-purchase relationships). In this paper, we explore two objectives: Objective 1: In-distribution Class Classification. Given the training and test nodes sharing the same distribution, where Ptrain(Gtrain) = Ptest(Gtest), and the conditional distribution Ptrain(y|Gtrain) = Ptest(y|Gtest), we aim to design … view at source ↗

**Figure 2.** TNT-OOD framework: 1) Structure encoders [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗

**Figure 3.** Embedding representation and OOD score visualisation of [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗

Figures from the paper (2 more)

**Figure 4.** Figure 4: LLM prompt for OOD class selection. LLM Prompt for LLM as OOD Detector on Label Shifts You are given a set of known categories indexed from 0 to [NUM_ID_CLASSES]-1, where they are in-distribution (ID) classes. For each test case, you will be given a ’Text’. The given s…

**Figure 5.** Figure 5: LLM prompt for LLM as Detector methods on label shifts. [PITH_FULL_IMAGE:figures/full_fig_p019_5.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

ALSA: Anchors in Logit Space for Out-of-Distribution Accuracy Estimation
cs.LG 2025-08 conditional novelty 6.0 of 10

ALSA learns anchors in logit space and uses their influence on unlabeled samples to estimate model accuracy under distribution shift.

Reference graph

Works this paper leans on

16 extracted references · 16 canonical work pages · cited by 1 Pith paper

[1]

You must select EXACTLY [NUM_OOD_CLASSES] classes as OOD

work page
[2]

id_classes

Provide your response as a JSON object with two keys: “id_classes” and “ood_classes”

work page
[3]

ins", "del

Thematically dissimilar labels to the ID labels. We used GPT-4o as the LLM model, giving the category names and descriptions (if available) as input. Text Shift. We provide further examples of text- level shifts involving synonym, antonym replace- ments as well as character level edits in Table 7. Domain-Based Sentiment Shift. For datasets with underlying...

work page arXiv 2024
[4]

In NeurIPS

GOOD: A graph out-of-distribution bench- mark. In NeurIPS. Yuxin Guo, Cheng Yang, Yuluo Chen, Jixi Liu, Chuan Shi, and Junping Du. 2023. A data-centric frame- work to endow graph neural networks with out-of- distribution detection ability. In KDD. David Ha, Andrew M. Dai, and Quoc V . Le. 2017. Hy- perNetworks. In ICLR. Dan Hendrycks and Kevin Gimpel. 201...

work page 2023
[7]

Out-of-distribution detection with deep nearest neighbors. In ICML. Yanran Tang, Ruihong Qiu, Yilun Liu, Xue Li, and Zi Huang. 2024a. Casegnn: Graph neural networks for legal case retrieval with text-attributed graphs. In ECIR. Yanran Tang, Ruihong Qiu, Hongzhi Yin, Xue Li, and Zi Huang. 2024b. Caselink: Inductive graph learning for legal case retrieval. ...

work page 2025
[10]

These approaches form the methodological foundation upon which specialised network-based detection methods build

measure deviation from class-conditional Gaussian distributions. These approaches form the methodological foundation upon which specialised network-based detection methods build. NLP OOD Detection. Recent OOD detec- tion techniques in NLP increasingly leverage PLM/LLMs to improve robustness on novel in- puts. Distance-based detectors applied to a model’s ...

work page 2022
[11]

NODESafe (Yang et al., 2024) extended this approach by refining extreme energy scores

introduced an energy-score based method with a propagation schema to effectively harness the intrinsic interdependencies between node fea- tures and network topology. NODESafe (Yang et al., 2024) extended this approach by refining extreme energy scores. GRASP (Ma et al., 2024) proposed an enhanced edge augmentation method to improve propagation effectiven...

work page 2024
[15]

Each key should contain an array of integers representing class indices (0-indexed)

work page

Show all 16 references

[16]

id_classes

Do not include any explanations, reasoning, or additional text. Example output format: {“id_classes”: [0, 1, 3, 5], “ood_classes”: [2, 4, 6]} Figure 4: LLM prompt for OOD class selection. LLM Prompt for LLM as OOD Detector on Label Shifts You are given a set of known categorie...

2018
[1998]

In ACM Digital Libraries

Citeseer: An automatic citation indexing sys- tem. In ACM Digital Libraries. Google. 2024. Gemini 2.5 flash. Large language model, available athttps://gemini.google.com/. Shurui Gui, Xiner Li, Limei Wang, and Shuiwang Ji

2024
[2017]

Mahalanobis distance-based approaches (Lee et al.,

utilises confidence scores from softmax out- puts, while ODIN(Liang et al., 2018) enhances this with temperature scaling and input perturbation. Mahalanobis distance-based approaches (Lee et al.,

2018
[2018]

In NeurIPS

A simple unified framework for detecting out- of-distribution samples and adversarial attacks. In NeurIPS. Haoyang Li, Xin Wang, Ziwei Zhang, and Wenwu Zhu. 2022a. Out-of-distribution generalization on graphs: A survey. CoRR. Yuhan Li, Peisong Wang, Xiao Zhu, Aochuan Chen, Hai...

2024
[2020]

In NeurIPS

Uncertainty aware semi-supervised learning on graph data. In NeurIPS. Tao Zou, Le Yu, Yifei Huang, Leilei Sun, and Bowen Du. 2023. Pretraining language models with text- attributed heterogeneous graphs. In EMNLP Find- ings. A Impact Statement Our work aims to inspire and pave ...

2023
[2021]

In NeurIPS

Graph posterior network: Bayesian predictive uncertainty for node classification. In NeurIPS. Yiyou Sun, Yifei Ming, Xiaojin Zhu, and Yixuan Li
[2022]

In EMNLP Findings

Holistic sentence embeddings for better out- of-distribution detection. In EMNLP Findings. Yuhan Chen, Yihong Luo, Yifan Song, Pengwen Dai, Jing Tang, and Xiaochun Cao. 2025. De- coupled graph energy-based model for node out-of- distribution detection on heterophilic graphs. I...

2025
[2023]

In NeurIPS

Dream the impossible: Outlier imagination with diffusion models. In NeurIPS. C. Lee Giles, Kurt D. Bollacker, and Steve Lawrence

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.

[1] [1]

You must select EXACTLY [NUM_OOD_CLASSES] classes as OOD

work page

[2] [2]

id_classes

Provide your response as a JSON object with two keys: “id_classes” and “ood_classes”

work page

[3] [3]

ins", "del

Thematically dissimilar labels to the ID labels. We used GPT-4o as the LLM model, giving the category names and descriptions (if available) as input. Text Shift. We provide further examples of text- level shifts involving synonym, antonym replace- ments as well as character level edits in Table 7. Domain-Based Sentiment Shift. For datasets with underlying...

work page arXiv 2024

[4] [4]

In NeurIPS

GOOD: A graph out-of-distribution bench- mark. In NeurIPS. Yuxin Guo, Cheng Yang, Yuluo Chen, Jixi Liu, Chuan Shi, and Junping Du. 2023. A data-centric frame- work to endow graph neural networks with out-of- distribution detection ability. In KDD. David Ha, Andrew M. Dai, and Quoc V . Le. 2017. Hy- perNetworks. In ICLR. Dan Hendrycks and Kevin Gimpel. 201...

work page 2023

[5] [7]

Out-of-distribution detection with deep nearest neighbors. In ICML. Yanran Tang, Ruihong Qiu, Yilun Liu, Xue Li, and Zi Huang. 2024a. Casegnn: Graph neural networks for legal case retrieval with text-attributed graphs. In ECIR. Yanran Tang, Ruihong Qiu, Hongzhi Yin, Xue Li, and Zi Huang. 2024b. Caselink: Inductive graph learning for legal case retrieval. ...

work page 2025

[6] [10]

These approaches form the methodological foundation upon which specialised network-based detection methods build

measure deviation from class-conditional Gaussian distributions. These approaches form the methodological foundation upon which specialised network-based detection methods build. NLP OOD Detection. Recent OOD detec- tion techniques in NLP increasingly leverage PLM/LLMs to improve robustness on novel in- puts. Distance-based detectors applied to a model’s ...

work page 2022

[7] [11]

NODESafe (Yang et al., 2024) extended this approach by refining extreme energy scores

introduced an energy-score based method with a propagation schema to effectively harness the intrinsic interdependencies between node fea- tures and network topology. NODESafe (Yang et al., 2024) extended this approach by refining extreme energy scores. GRASP (Ma et al., 2024) proposed an enhanced edge augmentation method to improve propagation effectiven...

work page 2024

[8] [15]

Each key should contain an array of integers representing class indices (0-indexed)

work page