Pith. sign in

REVIEW 5 major objections 5 minor 23 references

Context-Based Fake News Detection using Graph Based Approach: ACOVID-19 Use-case

T0 review · 5 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read Fake news leaves a structural trace in conceptual graphs, and the GBAD anomaly miner finds it in the paper's COVID-19 use-case.

desk verdict First GBAD-for-fake-news application, but no evaluation ties anomalies to ground truth, so the central detection claim is unsupported. read the letter →

arxiv 2507.13382 v1 pith:U5FKOWRQ submitted 2025-07-15 cs.CL cs.LG

Chandrashekar Muniyappa , Sirisha Velampalli This is my paper · ORCID
classification cs.CLcs.LG
keywords fakenewsdetectiongraph-basedanomalyGBADminimumdescriptionlengthconceptualgraphsCOVID-19misinformationnamedentityrecognitiontopicmodeling
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 is trying to establish that fake news has a structural signature that can be spotted without a trained classifier: once news articles are converted into conceptual graphs, false claims appear as anomalies from the corpus's dominant graph pattern. That would matter because anomaly mining on graphs is unsupervised at the discovery stage and returns a concrete subgraph, rather than a black-box score, explaining why an article looks suspicious. The authors take a public real/fake news corpus, add recent COVID-19 articles, narrow the collection to a single topic, and use sentence embeddings to pick the ten fake articles most similar to genuine COVID-19 news. Each of the three GBAD variants then finds a normative pattern and a deviant pattern, and the deviations match the curated false claims: vaccine approved versus vaccine rejected, president not infected versus president infected, and government versus lawmakers criticizing Sinovac. The paper reports this as the first application of the GBAD algorithm to the news domain.

What carries the argument

The machinery is the MDL-based GBAD system operating on a hand-designed conceptual-graph topology. The objective $M(S,G) = \mathrm{DL}(G \mid S) + \mathrm{DL}(S)$ selects the substructure $S$ whose description length, after compressing the full graph $G$, is smallest; that substructure is declared the norm. GBAD-MDL then looks for slight modifications of the norm, GBAD-P for improbable extensions or inserted parts, and GBAD-MPS for missing vertices or edges. The graph topology: News to in-line to Person, Organization, Location, Verb, and Noun, with entities and sentiment-verb tokens as leaves, is what turns a raw article into the data on which compression operates, so the entire method depends on that schema making genuine claims structurally common and false claims structurally different.

What would settle it

Run GBAD on the paper's corpus with the real/fake labels held out, count how many flagged anomalies are actually fake and how many fake articles are never flagged, and check whether the numbers beat random assignment; the paper reports no such label comparison.

Watch

Extended reading notes

Core claim

The central claim is that a false news article can be recognized by its shape after parsing into a fixed graph schema. In this paper's topology each article becomes a root News node with in-line slots for Person, Organization, Location, Verb, and Noun; values are filled by named-entity recognition and part-of-speech tagging. The GBAD system first compresses the whole set of instance graphs under the MDL principle to find the best normative substructure, then its three variants search for modifications, insertions, and deletions relative to that norm. The paper's reported result is that the anomalous substructures recovered this way are precisely the known false COVID-19 claims in the curated set: a positive/negative flip in the vaccine verdict, a negation of the president's infection, and a substitution of the actor criticizing Sinovac. The claim is not that the algorithm knows 'fake' in advance; it is that structural anomaly and fake news coincide for this use-case.

Load-bearing premise

The load-bearing premise is that every fake article produces a graph-level structural deviation under this hand-built Person, Organization, Location, Verb, and Noun schema, while genuine articles follow the common pattern; the paper never validates that premise against the dataset's real/fake labels.

Editorial extensions

If this is right

  • The same pipeline can be pointed at any unlabeled article collection and will return the corpus's dominant claim pattern plus the articles that deviate from it, without needing per-domain training data.
  • The three GBAD variants correspond to three distinct misinformational operations: a falsified version, an invented added element, and a removed or missing element, so flagged articles can be grouped by how they lie.
  • Anomalies are explainable as subgraphs, so a reader can see, for instance, 'vaccine got approved' and compare it with the normative 'vaccine rejected' rather than trusting an opaque score.
  • The authors' stated next steps: additional topics, heterogeneous data sets, and non-English articles, follow directly because neither the topology nor the MDL principle is tied to COVID-19 vocabulary.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Beyond the paper's three examples, a natural test would be to see whether GBAD still separates fake from real when the false claim is a fabricated entity rather than an inverted verdict, since all three reported anomalies are polarity flips or actor substitutions.
  • Because the corpus is curated before the anomaly search, a label-based precision/recall comparison against the public dataset would be needed to separate the claim 'fake news is graph-anomalous' from the weaker fact that the curated fake articles happen to be unusual.
  • The approach is more precisely an outlier-discovery tool: anything rare in graph structure gets flagged, so a corpus containing truthful but unusual events would presumably produce false positives; balanced data with rare truthful claims would test this directly.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 5 minor

Summary. The paper proposes an unsupervised graph-based approach to fake news detection. News articles are converted into conceptual graphs using NLP tools (POS tagging and NER), with a hand-crafted topology involving Person, Organization, Location, Verb, and Noun nodes. The GBAD family of algorithms (MDL, P, MPS) is then applied to discover a normative pattern and anomalous substructures. The authors claim that the anomalies correspond to fake news, and they illustrate this with three examples in Section 7 (e.g., normative pattern 'vaccine rejected' vs. anomalous pattern 'vaccine approved'). The experiments use a Kaggle real/fake news dataset augmented with NYT articles and COVID-19-related fake news, with one manually selected topic ('Politics') and one subtopic ('Corona Virus').

Significance. If the central claim were substantiated, the paper would offer a novel unsupervised tool for fake news detection, which could be valuable given the scarcity of labeled data in this domain. The use of GBAD, previously applied in domains like network security and activity recognition, to news text is a plausible transfer. However, the paper provides no quantitative evaluation linking the discovered anomalies to ground-truth fake news labels. The lack of precision, recall, F1, or any baseline comparison means that the significance of the approach cannot be assessed. The three examples in Section 7 are interpreted post hoc and are not validated against the labeled dataset. The potential strength of the unsupervised setting is therefore entirely unrealized in the present manuscript.

major comments (5)
  1. [Section 7, Figures 5-10] The central claim that anomalies found by GBAD correspond to fake news is unsupported. The only evidence is the three examples in Section 7, where the authors contrast the normative and anomalous patterns (e.g., vaccine rejected vs. vaccine approved) and assert that the anomalous pattern is fake. No ground-truth labels from the Kaggle dataset are used to verify that these anomaly-instances are actually fake, nor is there any analysis of how many anomalies were found and how many were correct. Without precision, recall, false-positive rate, or a comparison against a non-fake control set, the claim that GBAD detects fake news remains a postulate rather than a demonstrated result.
  2. [Section 4, Section 5] The evaluation is vulnerable to selection bias due to several unconstrained choices. The graph topology in Figure 2 is said to provide 'the best results' (Section 4), but 'best' is never defined with respect to any labeled metric. The topic and subtopic are manually selected ('Politics', then 'Corona Virus'), and only the top 10 fake news articles by embedding similarity to real news are retained (Section 5). These choices could have been iterated until a few plausible anomalies emerged. The paper does not report the number of topologies or parameter settings tried, nor does it provide any sensitivity analysis, so the reported examples cannot be taken as representative of the method's performance.
  3. [Section 5, Section 7] The evaluation is circular in an operational sense: the fake news candidates are selected because they are semantically similar to real news (via the Universal Sentence Encoder inner product), and then the structural anomalies found by GBAD in the resulting graph dataset are labeled as fake news. This design does not establish that structural deviation is a reliable indicator of falsity. In particular, it fails to rule out the possibility that real news articles also produce structural anomalies, or that some fake news articles are structurally normative. The paper provides no counterfactual analysis or control experiment with real-news-only graphs.
  4. [Section 3, Section 6] The method is not reproducible from the manuscript. Section 3 describes the GBAD algorithms at a high level but does not specify the implementation details used in the experiments: the MDL formula in Eq. (1) is generic, and no values are given for the beam width, search depth, or the threshold for 'slightly deviate' (Section 3.1). The input format is illustrated in Table 1, but the exact preprocessing steps (e.g., how POS tags map to the fixed topology, how graph instances are aggregated across articles) are not specified. Without these details, the experimental results in Section 6 and Section 7 cannot be reproduced or independently verified.
  5. [Section 2, Section 8] The literature positioning is incomplete and the contribution is overstated relative to prior work. The paper claims to be the first to apply GBAD to fake news detection, but Section 2 already cites numerous graph-based fake news detection methods (SAFER, GS2F, heterogeneous GNNs, temporal anomaly detection). The novelty over these works is not clearly articulated beyond the specific use of GBAD. Moreover, the conclusion (Section 8) does not mention any quantitative outcome or limitation regarding the evaluation gap; it simply states future work on varied domains and languages, which underscores the preliminary nature of the study.
minor comments (5)
  1. [Abstract and Section 1] The paper contains several grammatical and typographical errors, e.g., 'Its a significant concern' (Abstract) and 'varied domains and mine interesting patterns... language scan be a good future study' (Section 8). These should be corrected.
  2. [Section 4, Figure 2] Figure 2 (the proposed graph topology) and Table 1 (sample instance) are not clearly referenced in the text; the topology is described only loosely as 'Person, Organisation, Location etc.' and the mapping from the sample instantiation to the figure is not explained. A precise definition of the graph schema would improve clarity.
  3. [Section 5] The description of the data pipeline is incomplete: the number of NYT articles added, the size of the final graph dataset (though 3,602 instances are mentioned in Section 6), the balance between real and fake articles, and the criteria for 'similar' in the top-10 selection are not stated.
  4. [Section 6, Figure 4] Figure 4 shows running times but the unit of time is not specified on the axis label, and there is no explanation of why GBAD-P is slower; a brief interpretation would help.
  5. [References] Several references are malformed or incomplete, e.g., [4] has a title mismatch with the described work ('Detecting vehicular patterns' vs. the bird population study), [5] lists 'Lawrence B.' instead of a surname, and [13] has an inconsistent author list (Dong Zhou vs. Dong Zhou et al.). The references should be carefully checked against the cited works.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: the GBAD anomaly-detection computation is independent of the fake-news labels; the paper's weakness is lack of validation, not circularity.

full rationale

The paper's derivation chain is not circular in the sense of the seven enumerated patterns. GBAD-MDL, GBAD-P, and GBAD-MPS (Section 3) compute normative and anomalous substructures from graph-encoded news articles using the MDL objective M(S,G)=DL(G|S)+DL(S). This unsupervised computation is independent of any fake-news ground truth; the resulting anomalous patterns are genuinely data-driven outputs of the algorithm, not quantities defined in terms of the target label. The Section 7 statements that call certain anomalous patterns 'fake news found by GBAD-MDL' are post-hoc semantic interpretations of the algorithm's output, and the paper never operationalizes an independent link between structural anomalies and the Kaggle fake-news labels. That is a serious validation gap, and the central claim that anomalies equal fake news is unsupported without precision/recall or baseline comparisons, but unsupported is not the same as circular. The two self-citations to Velampalli et al. ([4], [9]) are related-work examples of GBAD applied in other domains; they are not load-bearing premises for the present result, because GBAD itself is attributed to the external SUBDUE system and MDL literature. The data-selection step (Section 5) curates the top 10 fake articles by embedding similarity to real articles, which may bias the examples, but it does not force the specific anomalous substructures (e.g., 'vaccine approved' versus 'vaccine rejected') to appear by construction. Therefore the appropriate finding is no significant circularity, with the caveat that the empirical claim remains unvalidated.

Assumptions & free parameters 3 free parameters · 3 assumptions · 0 invented entities

The central claim rests on a hand-designed graph schema and several manual filtering choices. No new physical or conceptual entities are introduced, but the graph topology and topic selection are free parameters fitted to make the approach yield interpretable results.

free parameters (3)
  • Number of LDA topics (15), subtopics (10), and manual topic selection = 15, 10, Politics/Corona Virus
    The authors chose 15 topics in LDA topic modeling, then 10 subtopics, and manually selected the Politics and Corona Virus topics (Section 5). These choices determine which articles enter the experiment.
  • Top-10 fake news selection threshold = 10
    After computing sentence embedding similarities, only the 10 most similar fake news articles were selected for graph construction (Section 5). This hand-picked subset biases the evaluation.
  • Graph topology = Fixed schema (News, Person, Organization, Location, Verb, Noun)
    Section 4 states the topology was selected because it 'provides the best results', but no evaluation criterion is given.
assumptions (3)
  • domain assumption LDA topic modeling and Universal Sentence Encoder embeddings produce semantically meaningful groupings
    The filtering of articles to the Corona Virus topic and the selection of the top-10 similar fake news articles rely on these models capturing semantic similarity (Section 5).
  • domain assumption POS and NER tagging by Spacy correctly identifies relevant tokens
    Graph construction depends on extracting Person, Organization, Location, Verb, and Noun tags from articles (Section 5).
  • ad hoc to paper MDL-based GBAD identifies normative and anomalous substructures that correspond to real and fake news
    The core claim assumes that structural anomalies in the conceptual graph are equivalent to fake news; this equivalence is presented in Section 7 without validation against labels.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Context-Based Fake News Detection using Graph Based Approach: ACOVID-19 Use-case." pith.science (2026). https://pith.science/paper/U5FKOWRQ

@misc{pith2026250713382,
  author       = {Pith},
  title        = {Pith review of: Context-Based Fake News Detection using Graph Based Approach: ACOVID-19 Use-case},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/U5FKOWRQ}},
  note         = {Machine review of arXiv:2507.13382}
}
read the original abstract

In today\'s digital world, fake news is spreading with immense speed. Its a significant concern to address. In this work, we addressed that challenge using novel graph based approach. We took dataset from Kaggle that contains real and fake news articles. To test our approach we incorporated recent covid-19 related news articles that contains both genuine and fake news that are relevant to this problem. This further enhances the dataset as well instead of relying completely on the original dataset. We propose a contextual graph-based approach to detect fake news articles. We need to convert news articles into appropriate schema, so we leverage Natural Language Processing (NLP) techniques to transform news articles into contextual graph structures. We then apply the Minimum Description Length (MDL)-based Graph-Based Anomaly Detection (GBAD) algorithm for graph mining. Graph-based methods are particularly effective for handling rich contextual data, as they enable the discovery of complex patterns that traditional query-based or statistical techniques might overlook. Our proposed approach identifies normative patterns within the dataset and subsequently uncovers anomalous patterns that deviate from these established norms.

Figures

Figures reproduced from arXiv: 2507.13382 by the authors.

Figure 1
Figure 1. Illustration of Different Types of Structural Anomalies Detected by GBAD [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Proposed Graph Topology The GBAD system employs three distinct algorithms to detect anomalies: 3.1 GBAD-MDL The GBAD-MDL algorithm identifies the best substructure in the graph using the MDL principle. Once the optimal pattern is identified, the algorithm searches for substructures that deviate slightly from this normative pattern, signaling potential anomalies. This approach is particularly effective in detecting m… view at source ↗
Figure 3
Figure 3. Topics of Interest: Topic Modeling semantic relationship between words in a given context when compared to the traditional word frequency approaches. In addition to CBOW, Mikolov et al., also proposed Continuous Skip-gram model, both of them fall under word2Vec (Word to Vector) architecture mainly used to measure the semantic relationship between words. All these new inventions in word embeddings gave birth to many … view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: Running times needed to discover patterns using GBAD [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Normative Pattern Found by GBAD-MDL 7 RESULTS AND DISCUSSION The GBAD algorithms discover anomalous substructures in the graph-representation of the provided data. • Normative pattern and anomalous pattern found by GBAD-MDL algorithm can be seen in Figures 5, 6.From th…
Figure 6
Figure 6. Figure 6: Anomalous Pattern Found by GBAD-MDL [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: Normative Pattern Found by GBAD-P [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 8
Figure 8. Figure 8: Anomalous Pattern Found by GBAD-P 8 [PITH_FULL_IMAGE:figures/full_fig_p008_8.png]
Figure 9
Figure 9. Figure 9: Normative Pattern Found by GBAD-MPS [PITH_FULL_IMAGE:figures/full_fig_p009_9.png]
Figure 10
Figure 10. Figure 10: Anomalous Pattern Found by GBAD-MPS 8 CONCLUSION AND FUTURE WORK In this work, we used graph-based approach to discover contextual fake news articles. Using NLP techniques, we are able to convert news articles to conceptual graphs. In this work, we analyzed only artic…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

23 extracted references · 15 canonical work pages

  1. [1]

    Gonzalez, Lawrence B

    Jesus A. Gonzalez, Lawrence B. Holder, and Diane J. Cook. 2002. Graph-Based Relational Concept Learning. In Proceedings of the Nineteenth International Conference on Machine Learning (ICML ’02). Morgan Kaufmann Publishers Inc., San Francisco, CA, USA, 219–226

    work page 2002

  2. [2]

    and Matthee, M

    De Beer, D. and Matthee, M. 2020. Approaches to identify fake news: A systematic literature review. Lecture Notes in Networks and Systems, pp. 13–22. doi:10.1007/978-3-030-49264-9_2

    work page doi:10.1007/978-3-030-49264-9_2 2020

  3. [3]

    Xinyi Zhou and Reza Zafarani. 2020. A Survey of Fake News: Fundamental Theories, Detection Methods, and Opportunities. ACM Comput. Surv. 53, 5, Article 109 (September 2021), 40 pages. https://doi.org/10.1145/3395046

    doi:10.1145/3395046 2020

  4. [4]

    and Eberle, W

    Velampalli, S., Mookiah, L. and Eberle, W. 2017. Detecting vehicular patterns using a graph-based approach. 2017 IEEE Conference on Visual Analytics Science and Technology (VAST), pp. 209–210. doi:10.1109/vast.2017.8585667. 9 CSAIDE 2025, March 07–09, 2025, Kuala Lumpur, Malaysia Muniyappa et al

    work page arXiv 2017

  5. [5]

    Paudel, William Eberle, and Lawrence B

    Ramesh C. Paudel, William Eberle, and Lawrence B. 2018. Anomaly Detection of Elderly Patient Activities in Smart Homes using a Graph-Based Approach. https://api.semanticscholar.org/CorpusID:219955427

    work page 2018

  6. [6]

    Chandra, Shantanum Mishra, Pushkar, Yannakoudakis, Helen, Shutova, and Ekaterina. 2020. ArXiv.org e-Print archive. https://doi.org/10.48550/arXiv.2008.06274

    work page Pith review arXiv doi:10.48550/arxiv.2008.06274 2020

  7. [7]

    Ketkar, Lawrence B

    Nikhil S. Ketkar, Lawrence B. Holder, and Diane J. Cook. 2005. Subdue: compression-based frequent pattern discovery in graph data. In Proceedings of the 1st international workshop on open source data mining: frequent pattern mining implementations (OSDM ’05). Association for Computing Machinery, New York, NY, USA, 71–76. https://doi.org/10.1145/1133905.1133915

    work page arXiv 2005

  8. [8]

    Rissanen, J. 2005. Minimum description length principle. Encyclopedia of Statistical Sciences [Preprint]. doi:10.1002/0471667196.ess1641.pub2

    work page doi:10.1002/0471667196.ess1641.pub2 2005

Show all 23 references
  1. [9]

    and Saxena, A

    Velampalli, S. and Saxena, A. 2019. A graph-based approach for IP network analysis. Lecture Notes in Networks and Systems, pp. 195–202. doi:10.1007/978-981-13-7150-9_20

    2019 doi

  2. [10]

    Tomas Mikolov, Kai Chen, Greg Corrado, and Jeffrey Dean. 2013. Efficient estimation of word representations in vector space. arXiv preprint, https://doi.org/10.48550/arXiv.1301.3781

    2013 doi arXiv

  3. [11]

    Blei, Andrew Y

    David M. Blei, Andrew Y. Ng, and Michael I. Jordan. 2003. Latent dirichlet allocation. J. Mach. Learn. Res. 3, 993–1022

    2003

  4. [12]

    D. Cer, Y. Yang, S. Kong, N. Hua, N. Limtiaco, R. St. John. 2018. Universal sentence encoder. arXiv preprint. https://doi.org/10.48550/arXiv.1803.11175

    2018 doi arXiv

  5. [13]

    Dong Zhou, Qiang Ouyang, Nankai Lin, Yongmei Zhou, and Aimin Yang. 2025. GS2F: Multimodal Fake News Detection Utilizing Graph Structure and Guided Semantic Fusion. ACM Trans. Asian Low-Resour. Lang. Inf. Process. 24, 2, Article 11 (February 2025), 22 pages. https://doi.org/10....

    2025 doi

  6. [14]

    and Mohana

    Hiremath, P., Kalagi, S.S. and Mohana. 2023. Analysis of fake news detection using graph neural network (GNN) and Deep Learning. 2nd International Conference on Automation, Computing and Renewable Systems (ICACRS). doi:10.1109/icacrs58579.2023.10405304

    2023

  7. [15]

    Bingbing Xie, Xiaoxiao Ma, Jia Wu, Jian Yang, Shan Xue, and Hao Fan. 2023. Heterogeneous Graph Neural Network via Knowledge Relations for Fake News Detection. In Proceedings of the 35th International Conference on Scientific and Statistical Database Management (SSDBM ’23). Ass...

    2023

  8. [16]

    Benamira, A. et al. 2019. Semi-supervised learning and Graph Neural Networks for fake news detection’. Proceedings of the 2019 IEEE/ACM International Conference on Advances in Social Networks Analysis and Mining, pp. 568–569. doi:10.1145/3341161.3342958

    2019

  9. [17]

    Teddy Lazebnik and Or Iny. 2024. Temporal graphs anomaly emergence detection: benchmarking for social media interactions. Applied Intelligence 54, 23 (Dec 2024), 12347–12356. https://doi.org/10.1007/s10489-024-05821-3

    2024 doi

  10. [18]

    Available at: https://www.kaggle.com/datasets/clmentbisaillon/fake-and-real-news- dataset

    Clmentbisaillon (2017) Fake-and-real-news-dataset, Kaggle. Available at: https://www.kaggle.com/datasets/clmentbisaillon/fake-and-real-news- dataset

    2017

  11. [19]

    Hezhe Qiao, Chaoxi Niu, Ling Chen, Guansong Pang. 2025. AnomalyGFM: Graph Foundation Model for Zero/Few-shot Anomaly Detection. arXiv. https://doi.org/10.48550/arXiv.2502.09254

    2025 doi arXiv

  12. [20]

    Lanjun Wang, Zehao Wang, Le Wu, and An-An Liu. 2024. Bots Shield Fake News: Adversarial Attack on User Engagement based Fake News Detection. In Proceedings of the 33rd ACM International Conference on Information and Knowledge Management (CIKM ’24). Association for Computing Ma...

    2024

  13. [21]

    Lin, Y. et al. 2025. NeurIPS poster UNIGAD: Unifying Multi-level graph anomaly detection. Available at: https://neurips.cc/virtual/2024/poster/93390

    2025

  14. [22]

    Wang, R. et al. 2025. Context correlation discrepancy analysis for graph anomaly detection. IEEE Transactions on Knowledge and Data Engineering, 37(1), pp. 174–187. doi:10.1109/tkde.2024.3488375

    2025

  15. [23]

    Dongcheng Zou, Hao Peng, and Chunyang Liu. 2024. A Structural Information Guided Hierarchical Reconstruction for Graph Anomaly Detection. In Proceedings of the 33rd ACM International Conference on Information and Knowledge Management (CIKM ’24). Association for Computing Machi...

    2024

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.