Estimating the Effective Topics of Articles and journals Abstract Using LDA And K-Means Clustering Algorithm

Aras M. Ismael; Karmand Hussein Abdalla; Shadikur Rahman; Umme Ayman Koana

REVIEW 4 major objections 3 minor 20 references

Estimating the Effective Topics of Articles and journals Abstract Using LDA And K-Means Clustering Algorithm

T0 review · 4 major / 3 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read The paper argues that combining LDA topic modelling with K-Means clustering and WordNet gives the most reliable keyphrase extraction from article and journal abstracts.

desk verdict Unreadable preprint with an unsupported central claim; desk reject until the authors supply a readable manuscript and a real evaluation. read the letter →

arxiv 2508.16046 v1 pith:KRLNYE6D submitted 2025-08-22 cs.IR

Shadikur Rahman , Umme Ayman Koana , Aras M. Ismael , Karmand Hussein Abdalla This is my paper · ORCID

classification cs.IR

keywords LDAK-MeansclusteringkeyphraseextractiontopicmodellingtextWordNetabstractanalysissearchstringconstruction

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

This paper tries to show that a pipeline built from Latent Dirichlet Allocation (LDA) topic modelling, K-Means clustering, and the WordNet lexical database can extract keyphrases from article and journal abstracts more reliably than alternative approaches. The authors position keyphrase extraction as a practical aid: researchers who need to build precise search strings from abstracts can use the automatically produced keyphrases to avoid misunderstandings caused by surface wording. The contribution is an applied combination of existing text-mining methods rather than a new algorithm, and the supporting evidence comes from running the pipeline on the authors' own corpus of abstracts.

What carries the argument

The central machinery is the combination of LDA (Latent Dirichlet Allocation, a generative probabilistic topic model), K-Means clustering (a partitional clustering method), and the WordNet lexical database. LDA assigns each document a distribution over latent topics, K-Means groups the documents into clusters based on their representations, and WordNet provides lexical relations used to organise or select candidate keyphrases. Together the three components turn a collection of abstracts into a short list of keyphrases that the authors treat as effective topical descriptions.

What would settle it

Take a gold-standard collection of abstracts with author-assigned keywords, run the LDA plus K-Means pipeline, and measure overlap between the extracted keyphrases and the assigned keywords. If the overlap is no better than a simple TF-IDF baseline, the paper's reliability claim would be contradicted.

Watch

Extended reading notes

Core claim

On its own corpus of article and journal abstracts, the paper claims that LDA and K-Means, supported by WordNet, achieve the most reliable performance for keyphrase extraction. The authors connect this technical result to a concrete use case: helping a researcher construct a search string from journals and articles without being misled by inconsistent terminology. In effect, the paper proposes that topic modelling and text clustering, which are usually treated as separate tasks, can be combined into one keyphrase extraction workflow that yields dependable descriptors for a set of abstracts.

Load-bearing premise

The reliability claim rests on treating cluster and topic coherence as a valid measure of keyphrase quality, rather than comparing the extracted keyphrases with known correct keyphrases or human judgments.

Editorial extensions

If this is right

Researchers could generate candidate search strings directly from a set of abstracts without reading every abstract in full.
The same LDA plus K-Means pipeline can be rerun on new corpora to produce topic labels and keyphrases automatically.
Adding a lexical database like WordNet may reduce mismatches caused by synonyms and variant phrasing in retrieved documents.
If the reliability claim holds, keyphrase extraction becomes feasible for large abstract collections where manual indexing is impractical.

Reading between the lines

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

A stronger evaluation would compare the extracted keyphrases against author-assigned keywords or human annotations; the paper's internal cluster and topic measurements alone do not separate the combined pipeline from simpler baselines.
The reliability claim likely depends on the homogeneity of the corpus; for mixed-domain abstract collections, the fixed topic and cluster counts would probably need adjustment.
If WordNet's coverage of specialised domain terms is limited, the lexical-database step could become the bottleneck, and a domain-specific dictionary might behave differently.
The same pipeline could be transferred from abstracts to full texts, where keyphrase quality may differ because full texts contain more sections and background material.

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.

Desk Editor's Note

read the letter

You asked me to look at arXiv:2508.16046. Bottom line: this is not a reviewable paper in its current form. The full text is almost entirely corrupted by an encoding issue, so the only substantive content is the abstract. And the abstract makes a comparative claim—LDA and K-Means achieve the “most reliable performance” for keyphrase extraction—that is supported by no visible evidence: no dataset, no hyperparameters, no evaluation metric, no baselines, no code, no numbers. I agree with the reader that the work is unverdictable, but I’d go further: even if the encoding were fixed, the claim as stated would need external grounding to mean anything. Internal cluster coherence or topic coherence does not show that extracted keyphrases match human or author-assigned ones; without that, any “reliability” verdict is circular with respect to the algorithms’ own outputs.

I’ll give credit where it’s due. The practical motivation—helping researchers build search strings from abstracts—is sensible, and the pipeline of LDA plus K-Means with WordNet is a plausible assembly of existing tools for that niche. But that’s a routine combination, not a new method, and the paper does not appear to compare against standard keyphrase extraction techniques or even cite them. So the novelty is low, and the claimed “most reliable” is unearned.

The soft spots are not subtle. The text corruption alone is enough to desk reject; the missing evaluation is the second, stronger reason. The reader’s weakest-assumption point is correct: the load-bearing premise is a valid measure of keyphrase quality, and that premise is not demonstrated anywhere in the recoverable text. The stress-test note holds up.

Who is this for? Possibly someone building a simple abstract-keyphrase tool for a bookkeeping task, but they would need to reimplement the pipeline from scratch and validate it themselves. That is not enough to justify referee time.

My recommendation: desk reject. If the authors resubmit with a readable manuscript, a clearly described dataset, and a comparison against at least one existing keyphrase extraction baseline using a standard metric like precision/recall/F1 against author keywords, then it becomes a marginal but possibly publishable workshop paper. As it stands, there is nothing to peer review.

Referee Report

4 major / 3 minor

Summary. The paper proposes a keyphrase-extraction pipeline for journal and article abstracts that combines LDA topic modeling, K-Means clustering, and the WordNet lexical database. The abstract claims that 'K-Means cluster and LDA algorithms achieve the most reliable performance for keyphrase extraction in our text documents' and asserts practical value for building search strings. In the submitted manuscript, however, the body text is almost entirely a corrupted encoding; only the abstract, some section headings, and fragments of equations or formulas are recoverable. No dataset, preprocessing details, hyperparameter values, evaluation metric, baseline comparison, or numerical result can be recovered. Consequently, the central comparative claim cannot be checked from the manuscript as submitted.

Significance. If the claimed result were established, the contribution would be a modest empirical demonstration that standard LDA and K-Means, augmented by WordNet, yield useful keyphrases on a particular abstract corpus. The paper does not provide machine-checked proofs, reproducible code, a recoverable experimental protocol, or falsifiable quantitative predictions. Because the manuscript contains no readable methods or results, and because the central assertion is an unquantified comparative claim, the significance cannot be assessed beyond the abstract's unsupported statement. The use of well-known algorithms is not itself a contribution without a validated evaluation.

major comments (4)

[Abstract; full text] The central claim, 'K-Means cluster and LDA algorithms achieve the most reliable performance for keyphrase extraction,' is not supported by any recoverable evidence. No quantitative results, baselines, evaluation metrics, or error bars are present. A comparative claim of 'most reliable' requires an evaluation against external ground truth or human judgment, not merely an internal demonstration of clustering or topic structure. This claim must either be substantiated with a reproducible benchmark or withdrawn.
[Sections 1-5 (passim)] The manuscript body is dominated by an encoding corruption that renders most of the text unreadable. I cannot verify the proposed method, the experimental setup, or any derivation. This is not a minor typographical issue: it prevents any substantive technical review. The authors need to resubmit a readable manuscript before the central claims can be evaluated.
[Method/Evaluation sections] No dataset is described: no corpus size, source, domain, or document selection criteria. Preprocessing steps, such as tokenization, stop-word removal, stemming, and lemmatization, are not recoverable. Hyperparameters, including the number of topics K for LDA, the number of clusters K for K-Means, and the keyphrase scoring threshold, are not specified. Without these, the experiments cannot be reproduced or assessed.
[Evaluation/Results (if present)] The evaluation appears to rely on internal outputs of the algorithms—topic top-terms or cluster centroids—which would make the 'reliability' claim self-referential. Cluster coherence or topic coherence does not demonstrate keyphrase extraction quality. The authors should compare extracted keyphrases against author-assigned keywords or human annotations, using metrics such as precision, recall, and F1.

minor comments (3)

[Title and Abstract] The title and abstract contain grammatical issues, e.g., 'journals Abstract' and 'has become a modern solution for the increasing number of text documents.' A careful language edit is needed.
[Section 3 (proposed method)] The role of WordNet is not clearly described. It is mentioned as a 'lexical database' for keyphrase extraction, but no detail is recoverable on how WordNet is integrated with LDA and K-Means. A precise description of the pipeline is needed.
[References] References, if any, are not recoverable from the corrupted text. The manuscript should include a complete reference list and proper citations for LDA, K-Means, WordNet, and related keyphrase-extraction work.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found; the central claim is empirically unsupported in the available text, but no claim reduces to its own inputs or to a self-citation chain.

full rationale

The supplied text is mostly encoding-corrupted, so the derivation chain is not recoverable. The only clear claims are in the abstract: LDA, K-Means, and WordNet were used for keyphrase extraction, and 'K-Means cluster and LDA algorithms achieve the most reliable performance for keyphrase extraction in our text documents.' This is a strong empirical claim, but no evaluation protocol, baseline comparison, numeric result, or fitted parameter is visible in the readable portions. A claim being unsupported by recoverable evidence is a correctness/validity concern, not circularity per the hard rules. There is no quoted equation showing that a predicted quantity is identical by construction to a fitted input, no self-citation invoked as load-bearing, and no imported uniqueness theorem. The absence of external gold-standard evaluation could make the 'most reliable performance' claim self-referential in the abstract, but the paper's visible text does not exhibit the specific reduction required to justify a circularity finding. Therefore the honest verdict is no significant circularity, score 0.

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

The paper introduces no new theoretical entities. It relies on standard LDA, K-Means, and WordNet assumptions, but the lack of disclosed parameter values and dataset details means the claimed reliability rests on unstated fitting choices.

free parameters (3)

Number of topics K (LDA)
LDA requires the number of topics K to be specified. The abstract and readable text do not state how K was chosen or whether it was tuned to the data, which affects the topic and keyphrase outputs.
Number of clusters K (K-Means)
K-Means requires the number of clusters to be set. This parameter directly determines the clustering structure and the resulting keyphrase groups, but its value or selection method is not disclosed.
Keyphrase scoring threshold
The pipeline uses WordNet for keyphrase extraction, which likely involves scoring or filtering candidate phrases. Any threshold or weighting is a free parameter that influences the reported performance.

assumptions (3)

domain assumption LDA bag-of-words assumption with Dirichlet priors on topic distributions
The paper uses LDA, which assumes documents are exchangeable bags of words and topics are generated from Dirichlet distributions. These assumptions may not match the structure of scientific abstracts.
domain assumption WordNet provides sufficient lexical coverage for scientific terms
WordNet is a general English lexical database. Scientific abstracts often contain specialized terminology and acronyms that may be absent or poorly represented in WordNet, which would affect keyphrase extraction.
domain assumption The chosen corpus is representative of the target domain
The paper does not describe the dataset or its selection criteria. The reliability claim is specific to 'our text documents' but is presented as generalizable, which requires the corpus to be representative of the intended use case.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Estimating the Effective Topics of Articles and journals Abstract Using LDA And K-Means Clustering Algorithm." pith.science (2026). https://pith.science/paper/KRLNYE6D

@misc{pith2026250816046,
  author       = {Pith},
  title        = {Pith review of: Estimating the Effective Topics of Articles and journals Abstract Using LDA And K-Means Clustering Algorithm},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KRLNYE6D}},
  note         = {Machine review of arXiv:2508.16046}
}

read the original abstract

Analyzing journals and articles abstract text or documents using topic modelling and text clustering has become a modern solution for the increasing number of text documents. Topic modelling and text clustering are both intensely involved tasks that can benefit one another. Text clustering and topic modelling algorithms are used to maintain massive amounts of text documents. In this study, we have used LDA, K-Means cluster and also lexical database WordNet for keyphrases extraction in our text documents. K-Means cluster and LDA algorithms achieve the most reliable performance for keyphrase extraction in our text documents. This study will help the researcher to make a search string based on journals and articles by avoiding misunderstandings.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

20 extracted references · 18 canonical work pages

[1]

@c L8G c( M;+Gd' | . ci NCQS62P␃ZH

11em plus .33em minus .07em 4000 4000 100 4000 4000 500 `\.=1000 = #1 \@IEEEnotcompsoconly \@IEEEcompsoconly #1 * [1] 0pt [0pt][0pt] #1 * [1] 0pt [0pt][0pt] #1 * \| ** #1 \@IEEEauthorblockNstyle \@IEEEcompsocnotconfonly \@IEEEauthorblockAstyle \@IEEEcompsocnotconfonly \@IEEEcompsocconfonly \@IEEEauthordefaulttextstyle \@IEEEcompsocnotconfonly \@IEEEauthor...

work page
[2]

Automatic label generation for news comment clus-ters

Ahmet Aker, Monica Paramita, Emina Kurtic, Adam Funk, Emma Barker, MarkHepple, and Rob Gaizauskas. Automatic label generation for news comment clus-ters. InProceedings of the 9th International Natural Language Generation Confer-ence, pages 61–69, 2016

work page 2016
[3]

Automatic Labelling of Topics with Neural Embeddings

Shraey Bhatia, Jey Han Lau, and Timothy Baldwin. Automatic labelling of topicswith neural embeddings.arXiv preprint arXiv:1612.05340, 2016

work page Pith review arXiv 2016
[4]

Latent dirichlet allocation.Journal of machine Learning research, 3(Jan):993–1022, 2003

David M Blei, Andrew Y Ng, and Michael I Jordan. Latent dirichlet allocation.Journal of machine Learning research, 3(Jan):993–1022, 2003

2003
[5]

Class-based n-gram models of natural language.Computationallinguistics, 18(4):467–479, 1992

Peter F Brown, Peter V Desouza, Robert L Mercer, Vincent J Della Pietra, andJenifer C Lai. Class-based n-gram models of natural language.Computationallinguistics, 18(4):467–479, 1992

work page 1992
[6]

Latent semantic analysis.Annual review of information scienceand technology, 38(1):188–230, 2004

Susan T Dumais. Latent semantic analysis.Annual review of information scienceand technology, 38(1):188–230, 2004

work page 2004
[7]

word2vec explained: deriving mikolov et al.’snegative-sampling word-embedding method.arXiv preprint arXiv:1402.3722, 2014

Yoav Goldberg and Omer Levy. word2vec explained: deriving mikolov et al.’snegative-sampling word-embedding method.arXiv preprint arXiv:1402.3722, 2014

arXiv 2014
[8]

Probabilistic latent semantic analysis

Thomas Hofmann. Probabilistic latent semantic analysis. InProceedings of the Fif-teenth conference on Uncertainty in artificial intelligence, pages 289–296. MorganKaufmann Publishers Inc., 1999

work page 1999

Show all 20 references

[9]

Polynomialtopic distribution with topic modeling for generic labeling

Syeda Sumbul Hossain, Md Rezwan Ul-Hassan, and Shadikur Rahman. Polynomialtopic distribution with topic modeling for generic labeling. InInternational Con-ference on Advances in Computing and Data Sciences, pages 409–419. Springer,2019

2019
[10]

Automaticlabelling of topic models

Jey Han Lau, Karl Grieser, David Newman, and Timothy Baldwin. Automaticlabelling of topic models. InProceedings of the 49th Annual Meeting of the Asso-ciation for Computational Linguistics: Human Language Technologies-Volume 1,pages 1536–1545. Association for Computational Lin...

2011
[11]

Best topicword selection for topic labelling

Jey Han Lau, David Newman, Sarvnaz Karimi, and Timothy Baldwin. Best topicword selection for topic labelling. InColing 2010: Posters, pages 605–613, 2010

2010
[12]

Learning the parts of objects by non-negativematrix factorization.Nature, 401(6755):788, 1999

Daniel D Lee and H Sebastian Seung. Learning the parts of objects by non-negativematrix factorization.Nature, 401(6755):788, 1999

1999
[13]

Some methods for classification and analysis of multivari-ate observations

James MacQueen et al. Some methods for classification and analysis of multivari-ate observations. InProceedings of the fifth Berkeley symposium on mathematicalstatistics and probability, volume 1, pages 281–297. Oakland, CA, USA, 1967

1967
[14]

Automatic la-beling of topics

Davide Magatti, Silvia Calegari, Davide Ciucci, and Fabio Stella. Automatic la-beling of topics. In2009 Ninth International Conference on Intelligent SystemsDesign and Applications, pages 1227–1232. IEEE, 2009

2009
[15]

Automatic labeling of multi-nomial topic models

Qiaozhu Mei, Xuehua Shen, and ChengXiang Zhai. Automatic labeling of multi-nomial topic models. InProceedings of the 13th ACM SIGKDD internationalconference on Knowledge discovery and data mining, pages 490–499, 2007

2007
[16]

Wordnet: a lexical database for english.Communications of theACM, 38(11):39–41, 1995

George A Miller. Wordnet: a lexical database for english.Communications of theACM, 38(11):39–41, 1995

1995
[17]

InFuture of Information and Communication Conference, pages224–236

Shadikur Rahman, Syeda Sumbul Hossain, Md Shohel Arman, Lamisha Rawshan,Tapushe Rabaya Toma, Fatama Binta Rafiq, and Khalid Been Md Badruzzaman.Assessing the effectiveness of topic modeling algorithms in discovering generic labelwith description. InFuture of Information and Co...

2020
[18]

Sharing clus-ters among related groups: Hierarchical dirichlet processes

Yee W Teh, Michael I Jordan, Matthew J Beal, and David M Blei. Sharing clus-ters among related groups: Hierarchical dirichlet processes. InAdvances in neuralinformation processing systems, pages 1385–1392, 2005

2005
[19]

Verbs semantics and lexical selection

Zhibiao Wu and Martha Palmer. Verbs semantics and lexical selection. InPro-ceedings of the 32nd annual meeting on Association for Computational Linguistics,pages 133–138. Association for Computational Linguistics, 1994

1994
[20]

Integrating document clustering and topic modeling.arXiv preprint arXiv:1309.6874, 2013

Pengtao Xie and Eric P Xing. Integrating document clustering and topic modeling.arXiv preprint arXiv:1309.6874, 2013

2013 arXiv

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.

[1] [1]

@c L8G c( M;+Gd' | . ci NCQS62P␃ZH

11em plus .33em minus .07em 4000 4000 100 4000 4000 500 `\.=1000 = #1 \@IEEEnotcompsoconly \@IEEEcompsoconly #1 * [1] 0pt [0pt][0pt] #1 * [1] 0pt [0pt][0pt] #1 * \| ** #1 \@IEEEauthorblockNstyle \@IEEEcompsocnotconfonly \@IEEEauthorblockAstyle \@IEEEcompsocnotconfonly \@IEEEcompsocconfonly \@IEEEauthordefaulttextstyle \@IEEEcompsocnotconfonly \@IEEEauthor...

work page

[2] [2]

Automatic label generation for news comment clus-ters

Ahmet Aker, Monica Paramita, Emina Kurtic, Adam Funk, Emma Barker, MarkHepple, and Rob Gaizauskas. Automatic label generation for news comment clus-ters. InProceedings of the 9th International Natural Language Generation Confer-ence, pages 61–69, 2016

work page 2016

[3] [3]

Automatic Labelling of Topics with Neural Embeddings

Shraey Bhatia, Jey Han Lau, and Timothy Baldwin. Automatic labelling of topicswith neural embeddings.arXiv preprint arXiv:1612.05340, 2016

work page Pith review arXiv 2016

[4] [4]

Latent dirichlet allocation.Journal of machine Learning research, 3(Jan):993–1022, 2003

David M Blei, Andrew Y Ng, and Michael I Jordan. Latent dirichlet allocation.Journal of machine Learning research, 3(Jan):993–1022, 2003

2003

[5] [5]

Class-based n-gram models of natural language.Computationallinguistics, 18(4):467–479, 1992

Peter F Brown, Peter V Desouza, Robert L Mercer, Vincent J Della Pietra, andJenifer C Lai. Class-based n-gram models of natural language.Computationallinguistics, 18(4):467–479, 1992

work page 1992

[6] [6]

Latent semantic analysis.Annual review of information scienceand technology, 38(1):188–230, 2004

Susan T Dumais. Latent semantic analysis.Annual review of information scienceand technology, 38(1):188–230, 2004

work page 2004

[7] [7]

word2vec explained: deriving mikolov et al.’snegative-sampling word-embedding method.arXiv preprint arXiv:1402.3722, 2014

Yoav Goldberg and Omer Levy. word2vec explained: deriving mikolov et al.’snegative-sampling word-embedding method.arXiv preprint arXiv:1402.3722, 2014

arXiv 2014

[8] [8]

Probabilistic latent semantic analysis

Thomas Hofmann. Probabilistic latent semantic analysis. InProceedings of the Fif-teenth conference on Uncertainty in artificial intelligence, pages 289–296. MorganKaufmann Publishers Inc., 1999

work page 1999