Pith. sign in

REVIEW 4 major objections 2 minor 68 references

GeoMoE: Divide-and-Conquer Motion Field Modeling with Mixture-of-Experts for Two-View Geometry

T0 review · 4 major / 2 minor · reviewed 2026-08-06 · deepseek-v4-flash

Pith's one-line read GeoMoE proposes a divide-and-conquer mixture-of-experts motion field model for two-view geometry and claims state-of-the-art pose/homography estimation; the supplied body text, though, is an unrelated Bengali topic-modeling paper.

desk verdict The submission is a different paper: the GeoMoE abstract is attached to a Bengali topic-modeling manuscript, so GeoMoE's claims have no supporting evidence. read the letter →

arxiv 2508.00592 v1 pith:GU2KKP3D submitted 2025-08-01 cs.CV

classification cs.CV
keywords GeoMoEtwo-viewgeometrymotionfieldestimationmixture-of-expertsrelativeposehomographyinlierprobabilitydivide-and-conquer
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's abstract proposes GeoMoE, a divide-and-conquer reformulation of two-view motion field estimation: instead of enforcing one global smoothness prior, it uses inlier-probability signals to decompose the motion field into heterogeneous sub-fields and then routes each sub-field through a dedicated expert in a mixture-of-experts rectifier. The intended claim is that this structure-aware decomposition plus expert routing decouples heterogeneous motion regimes, suppresses cross-sub-field interference and representational entanglement, and yields fine-grained motion-field rectification. On that basis the abstract reports that GeoMoE outperforms prior state-of-the-art methods in relative pose and homography estimation and shows strong generalization. The supplied full text, however, is not the GeoMoE paper: it presents an unrelated graph-based hybrid topic-modeling study for the Bengali language. Accordingly, the stated contribution is visible only in the abstract, and the submitted manuscript contains no derivation or experiment supporting it.

What carries the argument

The central machinery is the mixture-of-experts architecture applied to motion sub-fields, driven by two named components. The Probabilistic Prior-Guided Decomposition uses the matching process's inlier-probability signals as a prior to split the motion field into structure-aware sub-fields, so that outliers do not bias a single global estimate. The MoE-Enhanced Bi-Path Rectifier then processes each sub-field along two complementary paths—spatial-context and channel-semantic—and routes it to a customized expert, giving each heterogeneous motion regime its own rectification. The intended function is to decouple regimes and avoid representational entanglement across sub-fields; the design is described as minimalist, with routing and decomposition doing the work rather than a large monolithic network.

What would settle it

Confirm the manuscript's identity first: if the submitted full text contains a Bengali topic-modeling study and no geometric tables, the abstract's claims have no supporting evidence in the paper itself. To test the architectural claim after that, run GeoMoE on a public relative-pose or homography benchmark with depth discontinuities and viewpoint change, and compare against the prior methods the abstract says it outperforms; if the reported gains do not reproduce, or if removing the MoE routing yields no degradation, the divide-and-conquer claim fails.

Watch

Extended reading notes

Core claim

The GeoMoE abstract's central claim is that mixture-of-experts can implement divide-and-conquer over a two-view motion field: different spatial regions or motion regimes should not share a single rectification model. The paper introduces a Probabilistic Prior-Guided Decomposition that reads inlier-probability signals from the matching process and performs a structure-aware split of the motion field into heterogeneous sub-fields, curbing outlier-induced bias. Each sub-field is then handled by an MoE-Enhanced Bi-Path Rectifier, which refines the sub-field along a spatial-context path and a channel-semantic path before routing it to a customized expert. The claimed effect is decoupling of heterogeneous motion regimes, suppression of cross-sub-field interference and representational entanglement, and consequently more faithful motion fields, leading to state-of-the-art relative pose and homography estimation with strong generalization. Because the body text is an unrelated topic-modeling manuscript, this discovery is asserted rather than demonstrated in the submitted material.

Load-bearing premise

The load-bearing premise is that inlier-probability signals contain enough information about scene structure that splitting the motion field on those signals yields clean, correctly routed sub-fields without injecting new bias; a more basic but necessary premise is that the file submitted is actually the GeoMoE paper, which the current body text contradicts.

Editorial extensions

If this is right

  • If the abstract's claim is right, relative pose and homography estimation on scenes with extreme viewpoint and scale changes and depth discontinuities would improve over current state-of-the-art methods.
  • Inlier-probability decomposition would let matching confidences shape the motion estimate directly, reducing the influence of outliers on the final geometry.
  • Expert routing would give distinct motion regimes—foreground, background, planar, parallax—specialized rectification instead of a single global smoothness prior.
  • The 'minimalist design' claim implies the gains would come without heavy scene-specific architectural machinery, making the approach a compact drop-in for two-view pipelines.

Reading between the lines

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

  • If GeoMoE's decomposition works as advertised, the same inlier-probability split could serve as a confidence map for downstream tasks such as outlier rejection, uncertainty estimation, or iterative re-weighting in pose solvers; the abstract only uses it for routing.
  • A decisive test the paper does not report in the supplied material is systematic ablation of the routing: replacing the MoE with a single shared rectifier and measuring pose error on depth-discontinuity-heavy scenes would isolate whether the gains come from expert specialization or simply from the probabilistic decomposition.
  • Because the submitted full text is unrelated to the GeoMoE architecture, the public version of this page should be read as describing an intended contribution whose only available evidence is the abstract and, if accessible, the linked code release.
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

4 major / 2 minor

Summary. This submission, arXiv:2508.00592, presents an abstract for GeoMoE, a divide-and-conquer framework for two-view geometry that uses probabilistic prior-guided decomposition into motion sub-fields and an MoE-enhanced bi-path rectifier, claiming state-of-the-art relative pose and homography estimation. However, the supplied full text is not the GeoMoE paper: it is the GHTM paper, a graph-based hybrid topic model for Bengali, carrying its own arXiv identifier (2508.00605v2). The body contains no derivation, architecture, experiments, or error analysis for GeoMoE; all equations, tables, and results concern topic modeling. The central performance claim is therefore unverifiable from the submitted material.

Significance. The conceptual idea of using mixture-of-experts to handle heterogeneous motion sub-fields for two-view geometry is potentially interesting and, if substantiated, could be a meaningful contribution to the field. The manuscript as supplied, however, provides no evidence for this idea: there are no machine-checked proofs, no reproducible implementation within the text, no parameter-free derivations, and no falsifiable experimental predictions. The only concrete artifact is a GitHub URL, which cannot by itself support the performance claims. The significance of the claimed contribution cannot be assessed from the submitted manuscript.

major comments (4)
  1. [Abstract vs. full text] The submission's abstract describes GeoMoE for two-view geometry, but the full text is a different paper on Bengali topic modeling (GHTM). The word 'GeoMoE' never appears in the body, and no section or equation addresses relative pose estimation, homography estimation, motion fields, or mixture-of-experts. Thus the paper's central claim is entirely unsupported by the supplied manuscript.
  2. [GHTM method section ('Graph-based Hybrid Topic Model')] The abstract promises specific components, namely Probabilistic Prior-Guided Decomposition and MoE-Enhanced Bi-Path Rectifier, but the manuscript contains no equations, algorithms, or architectural descriptions for them. Equations (1)-(11) define TF-IDF weighting, GCN propagation, and NMF factorization for document embeddings; these are unrelated to motion field decomposition or two-view geometry.
  3. [Results and analysis] The experimental section reports topic coherence (NPMI), topic diversity (TD/IRBO), and runtimes on Bengali datasets and 20Newsgroups (Tables 7-10). There are no benchmarks for relative pose or homography estimation, no comparisons to prior two-view geometry methods, and no ablations of GeoMoE's components. The abstract's claim that GeoMoE 'outperforms prior state-of-the-art methods' is therefore not verifiable.
  4. [Reproducibility statement (Abstract)] The only GeoMoE artifact provided is a GitHub URL. The manuscript does not specify training datasets, evaluation protocols, metrics, or model configurations for two-view geometry, so the claimed results cannot be reproduced or checked from the submitted material.
minor comments (2)
  1. [General metadata] The title, abstract, and body describe different papers, and the arXiv identifier in the header (2508.00605v2) does not match the submission ID (2508.00592); this metadata inconsistency must be resolved.
  2. [Introduction of the GHTM text] The GHTM manuscript contains OCR artifacts, such as the dataset name rendered as a blank block in the Introduction, and would need thorough proofreading before any future resubmission.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity can be identified; the supplied full text is a different paper, so the GeoMoE claims are unverifiable but not circular.

full rationale

The GeoMoE abstract claims a derivation chain (Probabilistic Prior-Guided Decomposition, MoE-Enhanced Bi-Path Rectifier) and state-of-the-art pose and homography results, but the supplied full text is arXiv:2508.00605v2, a Bengali topic-modeling paper (GHTM) unrelated to GeoMoE. No GeoMoE equations, ablations, or benchmark tables are present, so there is no derivation chain to walk and no fitted parameter renamed as prediction can be exhibited. The word 'GeoMoE' never appears in the body text, and every equation (Eq. 1 through Eq. 11) concerns TF-IDF, GloVe, GCN, and NMF topic modeling. Because the hard rules require quoting a specific reduction (e.g., Eq. X equals Eq. Y by construction, or a fitted input relabeled as a prediction), the absence of the relevant manuscript text means no circular step can be substantiated. The mismatch is a severe correctness and reproducibility problem: the central state-of-the-art claim rests on unavailable evidence, but it is not circular, because the claim is unverified rather than equivalent to its inputs. Accordingly, the circularity score is 0, with the noted caveat that the GeoMoE contribution cannot be assessed from the provided material.

Assumptions & free parameters 0 free parameters · 2 assumptions · 0 invented entities

The central claim depends on unspecified hyperparameters (e.g., number of experts, routing assignments) and on domain assumptions about inlier-probability decomposition and MoE routing, but the supplied body contains no GeoMoE technical content to audit.

assumptions (2)
  • domain assumption Inlier probability signals support a structure-aware decomposition of the motion field into heterogeneous sub-fields.
    The abstract's Probabilistic Prior-Guided Decomposition assumes that grouping motion vectors by inlier probability yields sub-fields with distinct structure; no derivation is provided.
  • domain assumption Routing each sub-field to a dedicated Mixture-of-Experts module suppresses cross-sub-field interference and representational entanglement.
    The abstract asserts the MoE-Enhanced Bi-Path Rectifier achieves this decoupling, but no experimental or theoretical support appears in the supplied text.

how reviews work

0 comments
Cite this review

Pith. "Pith review of GeoMoE: Divide-and-Conquer Motion Field Modeling with Mixture-of-Experts for Two-View Geometry." pith.science (2026). https://pith.science/paper/GU2KKP3D

@misc{pith2026250800592,
  author       = {Pith},
  title        = {Pith review of: GeoMoE: Divide-and-Conquer Motion Field Modeling with Mixture-of-Experts for Two-View Geometry},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GU2KKP3D}},
  note         = {Machine review of arXiv:2508.00592}
}
read the original abstract

Recent progress in two-view geometry increasingly emphasizes enforcing smoothness and global consistency priors when estimating motion fields between pairs of images. However, in complex real-world scenes, characterized by extreme viewpoint and scale changes as well as pronounced depth discontinuities, the motion field often exhibits diverse and heterogeneous motion patterns. Most existing methods lack targeted modeling strategies and fail to explicitly account for this variability, resulting in estimated motion fields that diverge from their true underlying structure and distribution. We observe that Mixture-of-Experts (MoE) can assign dedicated experts to motion sub-fields, enabling a divide-and-conquer strategy for heterogeneous motion patterns. Building on this insight, we re-architect motion field modeling in two-view geometry with GeoMoE, a streamlined framework. Specifically, we first devise a Probabilistic Prior-Guided Decomposition strategy that exploits inlier probability signals to perform a structure-aware decomposition of the motion field into heterogeneous sub-fields, sharply curbing outlier-induced bias. Next, we introduce an MoE-Enhanced Bi-Path Rectifier that enhances each sub-field along spatial-context and channel-semantic paths and routes it to a customized expert for targeted modeling, thereby decoupling heterogeneous motion regimes, suppressing cross-sub-field interference and representational entanglement, and yielding fine-grained motion-field rectification. With this minimalist design, GeoMoE outperforms prior state-of-the-art methods in relative pose and homography estimation and shows strong generalization. The source code and pre-trained models are available at https://github.com/JiajunLe/GeoMoE.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

68 extracted references · 52 canonical work pages

  1. [1]

    TDAM: A Topic-Dependent Attention Model for Sentiment Analysis

    Pergola G, Gui L, He Y. TDAM: A Topic-Dependent Attention Model for Sentiment Analysis. Information Processing & Management. 2019;56(6):102084. Available from:https://doi.org/10.1016/j.ipm.2019. 102084. doi:10.1016/j.ipm.2019.102084

  2. [2]

    Learning Semantic Textual Similarity via Topic-informed Discrete Latent Variables

    Yu E, Du L, Jin Y, Wei Z, Chang Y. Learning Semantic Textual Similarity via Topic-informed Discrete Latent Variables. In: Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing (EMNLP). Association for Computational Linguistics (ACL); 2022. p. 4937-48. Available from: https://aclanthology.org/2022.emnlp-main.328/. doi:10.18...

  3. [4]

    Overview of the HASOC Track at FIRE 2024: Hate-Speech Identification in English and Bengali

    Raihan N, Ghosh K, Modha S, Satapara S, Gaur T, Dave Y, et al. Overview of the HASOC Track at FIRE 2024: Hate-Speech Identification in English and Bengali. In: Forum for Information Retrieval Evaluation (Working Notes)(FIRE 2024) December; 2024. p. 9-13. Available from:https://ceur-ws.org/Vol-4054/ T7-1.pdf

  4. [5]

    TOPIC MAP-BN: SCALABLE AND EXPLAINABLE FRAME- WORK FOR CROSS-SOURCE BANGLA NEWS RECOMMENDATION WITH BANGLABERT AND BERTOPIC

    Rahman MHH, Sunny SA. TOPIC MAP-BN: SCALABLE AND EXPLAINABLE FRAME- WORK FOR CROSS-SOURCE BANGLA NEWS RECOMMENDATION WITH BANGLABERT AND BERTOPIC. International Journal of Computer Science Engineering and Applications (IJCSEA). 2025 October;15(3/4/5):1-19. doi:10.5121/ijcsea.2025.15501

  5. [6]

    Latent Dirichlet Allocation

    Blei DM, Ng AY, Jordan MI. Latent Dirichlet Allocation. Journal of Machine Learning Research. 2003;3:993-1022. Available from:https://www.jmlr.org/papers/v3/blei03a.html

  6. [7]

    Learning the parts of objects by non-negative matrix factorization

    Lee DD, Seung HS. Learning the parts of objects by non-negative matrix factorization. Nature. 1999;401(6755):788–791. Available from:http://dx.doi.org/10.1038/44565. doi:10.1038/44565

  7. [8]

    Autoencoding Variational Inference for Topic Models

    Srivastava A, Sutton C. Autoencoding Variational Inference for Topic Models. In: Proceedings of the 5th International Conference on Learning Representations (ICLR). OpenReview.net; 2017. Available at https://arxiv.org/abs/1703.01488

  8. [9]

    Topic modeling in embedding spaces

    Dieng AB, Ruiz FJ, Blei DM. Topic modeling in embedding spaces. Transactions of the Association for Computational Linguistics. 2020;8:439-53

Show all 68 references
  1. [10]

    Pre-trainingisaHotTopic: ContextualizedDocumentEmbeddingsImprove Topic Coherence

    BianchiF,TerragniS,HovyD. Pre-trainingisaHotTopic: ContextualizedDocumentEmbeddingsImprove Topic Coherence. In: Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (V...

  2. [11]

    Cross-lingual Contextualized Topic Models with Zero- shot Learning

    Bianchi F, Terragni S, Hovy D, Nozza D, Fersini E. Cross-lingual Contextualized Topic Models with Zero- shot Learning. In: Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: Main Volume. Online: Association for Computat...

  3. [12]

    Top2Vec: Distributed Representations of Topics

    Angelov D. Top2Vec: Distributed Representations of Topics. arXiv preprint arXiv:200809470. 2020. Avail- able from: https://arxiv.org/abs/2008.09470

  4. [13]

    BERTopic: Neural topic modeling with a class-based TF-IDF procedure

    Grootendorst M. BERTopic: Neural topic modeling with a class-based TF-IDF procedure. arXiv preprint arXiv:220305794. 2022. Available from:https://arxiv.org/abs/2203.05794

  5. [14]

    GraphBTM: Graph Enhanced Autoencoded Variational Inference for Biterm Topic Model

    Zhu J, Jiang Y, Li Z, Zong C, Liu Q, Hovy E. GraphBTM: Graph Enhanced Autoencoded Variational Inference for Biterm Topic Model. In: Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing (EMNLP); 2018. p. 4663-72. Available from:https://aclantho...

  6. [15]

    Graph Contrastive Topic Model; 2023

    Luo Z, Liu L, Xie Q, Ananiadou S. Graph Contrastive Topic Model; 2023. arXiv preprint arXiv:2307.02078. Available from: https://arxiv.org/abs/2307.02078

  7. [16]

    GINopic: Topic Modeling with Graph Isomorphism Network

    Adhya S, Sanyal D. GINopic: Topic Modeling with Graph Isomorphism Network. arXiv preprint arXiv:240402115. 2025. Available from:https://arxiv.org/html/2404.02115

  8. [17]

    Graph2topic: An opensource topic modeling framework based on sentence embedding and community detection

    Zhang L, Liu C, Zhang J. Graph2topic: An opensource topic modeling framework based on sentence embedding and community detection. arXiv preprint arXiv:230406653. 2023

  9. [18]

    Topic Modeling Revisited: A Document Graph-based Neural Network Perspective

    Shen D, Qin C, Wang C, Dong Z, Zhu H, Xiong H. Topic Modeling Revisited: A Document Graph-based Neural Network Perspective. In: Advances in Neural Information Processing Systems (NeurIPS 2021). Curran Associates, Inc.; 2021. p. 1-13. Available from:https://proceedings.neurips....

  10. [19]

    TopicGPT: A Prompt-based Topic Modeling Framework

    Pham C, Hoyle A, Sun S, Resnik P, Iyyer M. TopicGPT: A Prompt-based Topic Modeling Framework. arXiv Preprint arXiv:231101449. 2024 Jan:2956–2984. Available from:https://doi.org/10.18653/v1/ 2024.naacl-long.164. doi:10.18653/v1/2024.naacl-long.164

  11. [20]

    Ethnologue: Languages of the World – Bengali; 2025

    SIL International. Ethnologue: Languages of the World – Bengali; 2025. Accessed: 2026-03-24.https: //www.ethnologue.com/language/ben/

  12. [21]

    Topic Modelling in Bangla Language: An LDA Approach to Optimize Topics and News Classification

    Helal MA, Mouhoub M. Topic Modelling in Bangla Language: An LDA Approach to Optimize Topics and News Classification. Computer and Information Science. 2018;11(4):77-7. Available from: https://www.researchgate.net/publication/328641108_Topic_Modelling_in_Bangla_Language_ An_LDA...

  13. [22]

    LDA2Vec: Combining LDA and Word2Vec for Topic Mod- eling in Bangla

    Hasan MM, Rahman MS, Islam MN. LDA2Vec: Combining LDA and Word2Vec for Topic Mod- eling in Bangla. International Journal of Computer Applications. 2019;177(28):1-7. Available from: https://www.ijcaonline.org/archives/volume177/number28/hasan-2019-ijca-919754.pdf. doi:10.5120/i...

  14. [23]

    Topic Modeling and Trend Analysis of Bengali News Articles

    Alam MS, Rahman MS, Islam MN. Topic Modeling and Trend Analysis of Bengali News Articles. Inter- national Journal of Computer Applications. 2020;176(34):1-7. Available from:https://www.ijcaonline. org/archives/volume176/number34/alam-2020-ijca-920000.pdf. doi:10.5120/ijca2020920000

  15. [24]

    Combining BERT with LDA: Improved Topic Mod- eling in Bengali Language

    Paul PC, Rahman M, Begum A, Ahmed MT. Combining BERT with LDA: Improved Topic Mod- eling in Bengali Language. IAENG International Journal of Computer Science. 2025;52(2):383-93. Available from:https://www.researchgate.net/publication/388779793_Combining_BERT_with_LDA_ Improved...

  16. [25]

    Likelihood Corpus Distribution: A Dirichlet-Polynomial Clustering Model for Bengali Topic Modeling

    Dawn S, Rahman MS, Islam MN. Likelihood Corpus Distribution: A Dirichlet-Polynomial Clustering Model for Bengali Topic Modeling. In: Proceedings of the 2024 International Conference on Natural LanguageProcessing(ICON);2024.p.100-10. Availablefrom: https://www.icon2024.org/proc...

  17. [26]

    Clustering LLM-based Word Embeddings to Determine Topics from Bangla Articles

    Rahman R, Ali ME. Clustering LLM-based Word Embeddings to Determine Topics from Bangla Articles. In: ACL Anthology; 2025. p. 309-21. Available from:https://aclanthology.org/2025.banglalp-1.25. pdf

  18. [27]

    Potrika: Raw and Balanced Newspaper Datasets in the Bangla Language with Eight Topics and Five Attributes

    Ahmad I, AlQurashi F, Mehmood R. Potrika: Raw and Balanced Newspaper Datasets in the Bangla Language with Eight Topics and Five Attributes. arXiv. 2022;abs/2210.09389. Available from:https: //arxiv.org/abs/2210.09389

  19. [28]

    Shironaam: Bengali News Headline Generation using Auxiliary Information

    Akash AU, Nayeem MT, Shohan FT, Islam T. Shironaam: Bengali News Headline Generation using Auxiliary Information. In: Vlachos A, Augenstein I, editors. Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics. Dubrovnik, Croat...

  20. [29]

    Bangla News Article Dataset (BNAD): A Standard Repository of 1.9 Million News Articles from Nine Bangla News Websites

    Saad AM, Mahi UN, Salim MS, Hossain SI. Bangla News Article Dataset (BNAD): A Standard Repository of 1.9 Million News Articles from Nine Bangla News Websites. Data in Brief. 2024;57:110874. Available from: https://doi.org/10.1016/j.dib.2024.110874. doi:10.1016/j.dib.2024.110874

  21. [30]

    BanFakeNews: A Dataset for Detecting Fake News in Bangla

    Hossain MZ, Rahman MA, Islam MS, Kar S. BanFakeNews: A Dataset for Detecting Fake News in Bangla. In: Proceedings of the Twelfth Language Resources and Evaluation Conference (LREC 2020). Marseille, France: European Language Resources Association (ELRA); 2020. p. 2862-71. Avail...

  22. [31]

    GloVe: Global Vectors for Word Representation

    Pennington J, Socher R, Manning C. GloVe: Global Vectors for Word Representation. In: Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP). Doha, Qatar: Association for Computational Linguistics; 2014. p. 1532-43. Available from: https...

  23. [32]

    Semi-Supervised Classification with Graph Convolutional Networks

    Kipf TN, Welling M. Semi-Supervised Classification with Graph Convolutional Networks. In: Proceedings of the 5th International Conference on Learning Representations (ICLR). OpenReview.net; 2017. Available at https://openreview.net/forum?id=SJU4ayYgl

  24. [33]

    Indexing by latent semantic analysis

    Deerwester S, Dumais ST, Furnas GW, Landauer TK, Harshman R. Indexing by latent semantic analysis. Journal of the American Society for Information Science. 1990;41(6):391–407. Available from:http://dx. doi.org/10.1002/(SICI)1097-4571(199009)41:6<391::AID-ASI1>3.0.CO;2-9 .doi:1...

  25. [34]

    Neuralvariationalinferenceandlearninginbeliefnetworks

    MnihA,GregorK. Neuralvariationalinferenceandlearninginbeliefnetworks. In: InternationalConference on Machine Learning. PMLR; 2014. p. 1791-9

  26. [35]

    Auto-encoding variational bayes

    Kingma DP, Welling M. Auto-encoding variational bayes. arXiv preprint arXiv:13126114. 2013

  27. [36]

    CluWords: exploiting semantic word clustering representation for enhanced topic modeling

    Viegas F, Canuto S, Gomes C, Luiz W, Rosa T, Ribas S, et al. CluWords: exploiting semantic word clustering representation for enhanced topic modeling. In: Proceedings of the twelfth ACM international conference on web search and data mining; 2019. p. 753-61

  28. [37]

    Large language models offer an alternative to the traditional approach of topic modelling

    Mu Y, Dong C, Bontcheva K, Song X. Large language models offer an alternative to the traditional approach of topic modelling. arXiv preprint arXiv:240316248. 2024 Mar. Available from:https://arxiv. org/abs/2403.16248

  29. [38]

    Addressing topic granularity and hallucination in large language models for topic modelling

    Mu Y, Bai P, Bontcheva K, Song X. Addressing topic granularity and hallucination in large language models for topic modelling. arXiv preprint arXiv:240500611. 2024 May. Available from:https://arxiv. org/abs/2405.00611

  30. [39]

    Mixing dirichlet topic models and word embeddings to make lda2vec

    Moody CE. Mixing dirichlet topic models and word embeddings to make lda2vec. arXiv preprint arXiv:160502019. 2016;(arXiv:1605.02019). Available from:http://arxiv.org/abs/1605.02019. doi:10.48550/arXiv.1605.02019

  31. [40]

    A Systematic Literature Review on English and Bangla Topic Modeling

    Ahmed MBU, Podder AA, Chowdhury MS, Al Mumin MA. A Systematic Literature Review on English and Bangla Topic Modeling. Journal of Computer Science. 2021;17(1):1-15. Available from: https://www.researchgate.net/publication/349973402_A_Systematic_Literature_Review_ on_English_and...

  32. [41]

    Bangla-BERT: Transformer- Based Efficient Model for Transfer Learning and Language Understanding

    Kowsher M, Sami AA, Prottasha NJ, Arefin MS, Dhar PK, Koshiba T. Bangla-BERT: Transformer- Based Efficient Model for Transfer Learning and Language Understanding. IEEE Access. 2022;10:91855-70. doi:10.1109/ACCESS.2022.3197662

  33. [42]

    Support vector machines and Word2vec for text classification with semantic features

    Lilleberg J, Zhu Y, Zhang Y. Support vector machines and Word2vec for text classification with semantic features. In: 2015 IEEE 14th International Conference on Cognitive Informatics & Cognitive Computing (ICCI*CC); 2015. p. 136-40.doi:10.1109/ICCI-CC.2015.7259377

  34. [43]

    Measuring document similarity with weighted averages of word embeddings

    Seegmiller B, Papanikolaou D, Schmidt LDW. Measuring document similarity with weighted averages of word embeddings. Explorations in Economic History. 2023;87:101494.doi:10.1016/j.eeh.2022.101494

  35. [44]

    Improving a tf-idf weighted document vector embedding; 2019

    Schmidt CW. Improving a tf-idf weighted document vector embedding; 2019. arXiv preprint. arXiv:1902.09875

  36. [45]

    Graph Convolutional Networks for Text Classification

    Yao L, Mao C, Luo Y. Graph Convolutional Networks for Text Classification. In: Proceedings of the AAAI Conference on Artificial Intelligence. vol. 33. AAAI; 2019. p. 7370-7

  37. [46]

    Graph Neural Networks: A Review of Methods and Applications

    Zhang S, Tong H, Xu J, Maciejewski R. Graph Neural Networks: A Review of Methods and Applications. ACM Computing Surveys (CSUR). 2020;52(4):1-37

  38. [47]

    A Comprehensive Survey on Graph Neural Networks

    Wu Z, Pan S, Chen F, Long G, Zhang C, Yu PS. A Comprehensive Survey on Graph Neural Networks. IEEE Transactions on Neural Networks and Learning Systems. 2021;32(1):4-24

  39. [48]

    Deep Graph Contrastive Representation Learning

    Zhu Y, Xu Y, Pan S, Hu R, Lin H, Zhou C. Deep Graph Contrastive Representation Learning. IEEE Transactions on Pattern Analysis and Machine Intelligence. 2022;44(12):8756-70. doi:10.1109/TPAMI.2021.3061165

  40. [49]

    Dual Graph Convolutional Networks for Graph-Based Semi-Supervised Classification

    Zhang M, Chen Y. Dual Graph Convolutional Networks for Graph-Based Semi-Supervised Classification. In: Advances in Neural Information Processing Systems (NeurIPS). vol. 33; 2020. p. 16697-707

  41. [50]

    Hybrid Margin Contrastive Loss for Graph Neural Networks

    Wu Z, Li H, Zhang P, Wang C. Hybrid Margin Contrastive Loss for Graph Neural Networks. Knowledge- Based Systems. 2022;246:108632.doi:10.1016/j.knosys.2022.108632

  42. [51]

    Cluster-GCN: An Efficient Algorithm for Training Deep and Large Graph Convolutional Networks

    Chiang WL, Liu X, Si S, Li Y, Bengio S, Hsieh CJ. Cluster-GCN: An Efficient Algorithm for Training Deep and Large Graph Convolutional Networks. In: Proceedings of the 25th ACM SIGKDD International ConferenceonKnowledgeDiscovery&DataMining.AssociationforComputingMachinery; 2019....

  43. [52]

    A Simple Framework for Contrastive Learning of Visual Representations

    Chen T, Kornblith S, Norouzi M, Hinton G. A Simple Framework for Contrastive Learning of Visual Representations. Proceedings of the 37th International Conference on Machine Learning. 2020:1597-607

  44. [53]

    Bangla SBERT - Sentence Embedding Using Multilingual Knowledge Distillation

    Uddin MS, Haque MA, Rifat RH, Kamal M, Gupta KD, George R. Bangla SBERT - Sentence Embedding Using Multilingual Knowledge Distillation. In: 2024 IEEE 15th An- nual Ubiquitous Computing, Electronics & Mobile Communication Conference (UEMCON); 2024. p. 495-500. Available from: h...

  45. [54]

    Visualizing data using t-SNE

    Van der Maaten L, Hinton G. Visualizing data using t-SNE. Journal of Machine Learning Research. 2008;9(11):2579-605

  46. [55]

    The Psycho-Biology of Language: An Introduction to Dynamic Philology

    Zipf GK. The Psycho-Biology of Language: An Introduction to Dynamic Philology. Cambridge, MA, USA: Houghton Mifflin; 1935

  47. [56]

    Lexical Diversity and Language Development

    Malvern D, Richards B, Chipere N, Durán P. Lexical Diversity and Language Development. Palgrave Studies in Language and Linguistics. London: Palgrave Macmillan; 2004

  48. [57]

    MTLD, vocd-D, and HD-D: A Validation Study of Sophisticated Approaches to Lexical Diversity Assessment

    McCarthy PM, Jarvis S. MTLD, vocd-D, and HD-D: A Validation Study of Sophisticated Approaches to Lexical Diversity Assessment. Behavior Research Methods. 2010;42(2):381-92.doi:10.3758/BRM.42.2.381

  49. [58]

    Cutting the Gordian Knot: The Moving-Average Type–Token Ratio

    Covington MA, McFall JD. Cutting the Gordian Knot: The Moving-Average Type–Token Ratio. Journal of Quantitative Linguistics. 2010;17(2):94-106. Available from:https://www.semanticscholar.org/paper/ 5fe88c0cd86d892593aa8f5f98039d9a290498e0. doi:10.1080/09296171003643098

  50. [59]

    Psychometric Evaluation of Lexical Diversity Indices: As- sessing Length Effects

    Fergadiotis G, Wright HH, Green SB. Psychometric Evaluation of Lexical Diversity Indices: As- sessing Length Effects. Journal of Speech, Language, and Hearing Research. 2015;58(3):840-52. doi:10.1044/2015_JSLHR-L-14-0280

  51. [60]

    Language and Thought

    Carroll JB. Language and Thought. Englewood Cliffs, NJ: Prentice-Hall; 1964

  52. [61]

    An Assessment of the Range and Usefulness of Lexical Diversity Measures and the Potential of the Measure of Textual Lexical Diversity (MTLD) [Ph.D

    McCarthy PM. An Assessment of the Range and Usefulness of Lexical Diversity Measures and the Potential of the Measure of Textual Lexical Diversity (MTLD) [Ph.D. dissertation]. University of Memphis; 2005

  53. [62]

    Twenty Newsgroups; 1997

    Mitchell T. Twenty Newsgroups; 1997. DOI: https://doi.org/10.24432/C5C323. UCI Machine Learning Repository

  54. [63]

    Automatic Evaluation of Topic Coherence

    Newman D, Lau JH, Grieser K, Baldwin T. Automatic Evaluation of Topic Coherence. In: Human Language Technologies: The 2010 Annual Conference of the North American Chapter of the Association for Computational Linguistics. Los Angeles, California: Association for Computational L...

  55. [64]

    Exploring the Space of Topic Coherence Measures

    Röder M, Both A, Hinneburg A. Exploring the Space of Topic Coherence Measures. In: Proceedings of the Eighth ACM International Conference on Web Search and Data Mining. ACM; 2015. p. 399-408. Available from: https://dl.acm.org/doi/10.1145/2684822.2685324. doi:10.1145/2684822.2685324

  56. [65]

    A Similarity Measure for Indefinite Rankings

    Webber W, Moffat A, Zobel J. A Similarity Measure for Indefinite Rankings. ACM Transactions on In- formation Systems. 2010;28(4):20:1-20:38. Available from:https://doi.org/10.1145/1852102.1852106. doi:10.1145/1852102.1852106

  57. [66]

    Software Framework for Topic Modelling with Large Corpora

    Řehůřek R, Sojka P. Software Framework for Topic Modelling with Large Corpora. In: Proceedings of the LREC 2010 Workshop on New Challenges for NLP Frameworks. Valletta, Malta: ELRA; 2010. p. 45-50. Available from: http://is.muni.cz/publication/884893/en

  58. [67]

    Terragni S, Galuzzi B, Tropeano P, Candelieri A, Archetti F, Fersini E. OCTIS: Comparing and Optimizing Topic Models is Simple! In: Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: System Demonstrations. Association f...

  59. [68]

    Convex and semi-nonnegative matrix factorizations

    Ding C, Li T, Jordan MI. Convex and semi-nonnegative matrix factorizations. IEEE Transactions on Pattern Analysis and Machine Intelligence. 2010;32(1):45-55.doi:10.1109/TPAMI.2008.277

  60. [69]

    Statistical Comparisons of Classifiers over Multiple Data Sets

    Demšar J. Statistical Comparisons of Classifiers over Multiple Data Sets. Journal of Machine Learning Research. 2006;7:1-30. Available from:https://jmlr.org/papers/v7/demsar06a/demsar06a.pdf

Pith tools

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