Pith. sign in

REVIEW 4 major objections 7 minor 34 references

Improving the Efficiency of Long Document Classification using Sentence Ranking Approach

T0 review · 4 major / 7 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read TF-IDF sentence ranking cuts long-document input in half with almost no accuracy loss.

desk verdict A plausible but overclaimed empirical study: TF-IDF sentence ranking helps for Marathi long-document classification, but the headline 0.33% accuracy drop is not trustworthy because the baseline and 100%-selection conditions don't match and the best config was chosen on the test set. read the letter →

arxiv 2506.07248 v2 pith:5AHVH2OH submitted 2025-06-08 cs.CL cs.LG

Prathamesh Kokate , Mitali Sarnaik , Manavi Khopade , Raviraj Joshi This is my paper
classification cs.CLcs.LG
keywords longdocumentclassificationsentencerankingTF-IDFcontextreductiontransformerefficiencyMarathinewsMahaBERTIndiclanguages
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 establish that a lightweight TF-IDF sentence ranking can replace most of a long document's text without meaningfully hurting classification accuracy. On the MahaNews long-document dataset of Marathi news articles, keeping only the top 40 percent of ranked sentences yields 94.39 percent accuracy, a 0.33 percent drop from the 94.706 percent full-context baseline. That reduction cuts inference latency by 43 percent with MahaBERT-v2, and ranked selection outperforms first, last, and random selection at small context sizes before all methods converge. The practical point is that input reduction can happen at the data level, before the transformer sees the text, rather than by changing the model architecture.

What carries the argument

The central object is the per-sentence TF-IDF score: each sentence $S_i$ is treated as a mini-document, term frequencies are computed within the sentence, inverse document frequencies are computed against all other sentences in the same article, and the sentence score is $\mathrm{Score}(S_i)=\sum_{t_j\in S_i}\mathrm{TF\text{-}IDF}(t_j)$. Top-ranked sentences are then selected either as a fixed count (1 to 5 sentences) or as a percentage of the document (10 to 100 percent). A length-normalized variant blends normalized TF-IDF with sentence length using $\mathrm{Score}=\lambda_1\mathrm{normTF\text{-}IDF}+\lambda_2\ell$ with $\lambda_1=1-\lambda_2$, letting the ranking trade term uniqueness against sentence length. This scoring carries the entire argument: it converts a long document into a short ranked list before the transformer ever processes the input.

What would settle it

Use a long-document corpus in which the label-bearing sentence consists entirely of common words while an unrelated sentence contains a rare technical term; if ranked selection consistently picks the rare-term sentence and falls below first-sentence selection, the TF-IDF informativeness assumption is falsified.

Watch

Extended reading notes

Core claim

The central claim is that sentence-level TF-IDF, computed by treating each sentence of an article as its own document and summing the TF-IDF weights of its terms, ranks sentences by label-relevant informativeness well enough that a standard BERT-style classifier needs only a fraction of the document. With fixed-count selection, five top-ranked sentences reach 94.19 percent accuracy, within 0.544 percent of the full-context baseline. With percentage-based selection, 40 percent coverage reaches 94.39 percent, a 0.33 percent drop from full context while inference latency falls by 43 percent. The paper also reports that ranked selection consistently orders as ranked greater than first, random, and last, and that adding sentence length through a weighted normalization term changes results only marginally in fixed-count mode. The conclusion is that a large part of a long document is redundant for classification, and a bag-of-words ranking captures enough of the informative core to preserve accuracy.

Load-bearing premise

The ranking assumes that summing a sentence's within-article TF-IDF scores measures how informative that sentence is for the article's category.

Editorial extensions

If this is right

  • A standard transformer can classify long documents accurately from a short ranked excerpt, so data-level context reduction is a viable complement to architectural changes.
  • At 40 percent coverage the method reaches 94.39 percent accuracy with 43 percent lower inference latency, making it practical where response time matters.
  • The consistent ordering rank greater than first, random, and last shows that where sentences come from matters more than how many are used, at least at small context sizes.
  • All selection methods converge as coverage approaches 100 percent, so the ranking advantage is concentrated exactly where the computational savings are largest.

Reading between the lines

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

  • Because the scoring is a bag-of-words heuristic, sentences that carry the label through pronouns, negation, or high-frequency phrasing may rank low; a testable extension is to rerank with a semantic model and measure whether the 40 percent accuracy point moves.
  • The 43 percent latency figure covers transformer inference only, not the preprocessing cost of tokenizing and scoring sentences; an end-to-end comparison would show where the practical break-even sits.
  • The same pipeline should transfer to other languages in the IndicNews collection whenever a sentence tokenizer and a language-specific BERT exist, though the paper evaluates only Marathi.
  • Percentage-based selection is likely to be more robust than fixed-count selection for documents of very different lengths, since it scales the retained context with document size.
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 / 7 minor

Summary. The paper proposes a TF-IDF-based sentence ranking method for long document classification, aiming to reduce input length for BERT-style models without modifying the architecture. The method ranks sentences by the sum of within-article TF-IDF scores and selects either a fixed number or a percentage of top-ranked sentences. A variant combines length-normalized TF-IDF scores with sentence length via a hyperparameter λ2. Experiments on the MahaNews LDC dataset (Marathi news articles) with MahaBERT-v2 compare ranked, first, last, and random selection at various counts and percentages. The paper claims near-identical accuracy to the full-context baseline (94.706%) with a 0.33% drop at 40% context, plus a 43% inference-latency reduction.

Significance. If the efficiency and accuracy-parity claims hold, the method offers a simple, model-agnostic, and practical alternative to architectural modifications for long document classification, particularly for low-resource languages like Marathi. The paper provides a broad comparison of selection strategies across multiple context budgets, and the directional ordering ranked > first > random > last is consistently reflected in the reported point estimates. The focus on a non-English dataset is a valuable contribution. However, the headline numerical claim (0.33% drop and 43% latency reduction) is not currently supported by the reported experimental protocol, because the full-context baseline is not reproduced by the same pipeline at 100% selection, the best configuration appears to be chosen post hoc from the test set, and no variance estimates are provided. These issues make the central efficiency claim unverifiable in its present form.

major comments (4)
  1. [Section 4.2 (Table 4) vs. Section 3.1] The full-context baseline is not reproduced by the same pipeline at 100% selection. Section 3.1 states the full-document accuracy is 94.706%, but Table 4 reports 100% selection accuracies of 94.11–94.78% across methods, with the Ranked Normalized condition (94.78%) exceeding the stated baseline. The claimed 0.33% drop at 40% context (94.39% vs. 94.706%) is therefore not a controlled comparison; using the matched 100% Ranked Normalized condition as baseline, the drop is 0.39 percentage points, and using other 100% conditions it ranges from 0.07 to 0.66 points. The authors should either use the 100% condition from the same pipeline as the baseline or provide repeated-run statistics demonstrating that 94.706% is within noise of the 100% conditions.
  2. [Section 4.2 (Tables 2 and 4)] The best result (40% selection, Ranked Normalized, λ2=0.7) is selected after inspecting test-set accuracies over a grid of percentages and λ2 values. The paper does not describe a validation-based configuration-selection procedure, despite a 2,548-sample validation split being mentioned in Section 3.1. Test-set-driven selection over a grid of roughly 50 configurations (10 percentages × 4 λ2 values plus non-normalized variants) makes the reported 94.39% an optimistic, post-hoc maximum. The authors should select the configuration on the validation split and report test accuracy only for that configuration, or otherwise quantify the selection bias, e.g., by reporting all grid results and applying a multiple-comparison correction.
  3. [Section 4 (Tables 1–4)] No standard deviations, confidence intervals, or repeated-seed experiments are reported. Differences of 0.1–0.5 percentage points (e.g., Table 4: 94.39% vs. 94.51% at 50%, or Table 1: 93.17% vs. 93.01% at 2 sentences) are used to support the ordering ranked > first > random > last and the near-parity claim. Without variance estimates, these differences may be within run-to-run noise. The authors should report means and standard deviations over at least 3–5 seeds, or provide statistical significance tests for the key comparisons, especially for the claimed 0.33% drop relative to the baseline.
  4. [Section 4.3 (Inference Time)] The 43% inference-latency reduction is reported without details of the measurement protocol: hardware, batch size, number of runs, and whether the TF-IDF ranking time is included or excluded. Since the ranking computation is an additional preprocessing step, its cost should be factored into the efficiency comparison. Without this information, the latency claim is not reproducible, and the practical efficiency advantage of the method is not fully demonstrated.
minor comments (7)
  1. [Section 3.2 (Eq. 1)] The summation in the score definition is garbled: "Score(Si) = X tj∈Si TF-IDF(ti)" should be "Score(S_i) = \sum_{t_j \in S_i} TF-IDF(t_j)". The same subscript error appears in the following lines.
  2. [Section 3.3 (Score formula)] The formula "Score = (λ1 · Normalized_TF_IDF) + (λ2 · length)" does not define "length" (number of tokens, characters, or words) nor the exact form of "Normalized_TF_IDF" (is it the total TF-IDF sum divided by length, or a per-token average?). Please specify the computation precisely.
  3. [Section 4.3] The phrase "only 0.33% below" should be "0.33 percentage points below" to avoid ambiguity between absolute and relative differences.
  4. [Section 4.2] The sentence "by reducing the context to just 40 to 50 percent ... we are still able to achieve an impressive accuracy of 94.39%" is imprecise: 94.39% corresponds specifically to 40% with the Ranked Normalized variant and λ2=0.7; at 50% the Ranked variant gives 94.51%. Please state the exact configuration to which the claim refers.
  5. [Section 2.1] "Exponential attention overhead" should be "quadratic attention complexity" to match the standard description of transformer self-attention.
  6. [Table 2 and Section 4.1] The claim "Accuracy peaks at λ2 = 0.7" holds for the 4-sentence case but is not uniform across all sentence counts (e.g., for 2 sentences λ2=0.2 gives 92.82% vs. λ2=0.7 gives 92.73%). Rephrase to avoid overgeneralization.
  7. [Section 3.2] The phrase "we propose a novel sentence ranking technique" is overstated since TF-IDF sentence ranking is a well-established approach. Please temper the novelty claim or explicitly differentiate the contribution (e.g., the within-article normalization scheme and the application to Marathi long-document classification).

Circularity Check

2 steps flagged · score 6.0 of 10

Headline 0.33%-drop result is a post-hoc test-grid optimum compared against a self-cited baseline that Table 4's own 100% runs do not reproduce; the underlying ranking-vs-baseline comparison remains empirical, so circularity is partial.

  1. fitted input called prediction [Section 4.1 (Normalization Results) and Section 4.2 (Data Percentage Approach); Tables 2 and 4]
    "Accuracy peaks at λ2 = 0.7, indicating that longer sentences are more informative in minimal contexts. […] Notably, at just 40% of the original document context, our method achieves 94.39% accuracy—only 0.33% below the full-context baseline—while reducing inference latency by 43%."

    The configuration producing the headline number is selected after observing test-set accuracies. Section 3.3 defines a tunable score, Score = (λ1 · Normalized_TF_IDF) + (λ2 · length), and Table 2 grids λ2 over {0.2, 0.5, 0.7, 1.0}, reporting that accuracy 'peaks' at λ2 = 0.7. Table 4 then grids coverage from 10% to 100%, and Section 4.2 picks the 40% row (94.39%) as the efficiency-accuracy sweet spot. No held-out validation, pre-registered configuration, or repeated-seed variance is reported. The claimed 0.33% drop is therefore the value of the best cell of a test-set grid, not an independent prediction; it is statistically forced by the selection procedure.

  2. self citation load bearing [Section 3.1 (Training and Testing) and Section 4.2; Table 4]
    "When trained and evaluated on the full LDC dataset, the MahaBERT model, fine-tuned on L3Cube-MahaCorpus and other public Marathi datasets, achieved an accuracy of 94.706%. […] Using the full-length documents for training and testing yields an accuracy of 94.706%, which serves as the baseline for comparison."

    The baseline used to compute the central accuracy-parity claim is inherited from the authors' prior L3Cube MahaBERT/MahaNews work, not reproduced in this paper's pipeline. Yet Table 4's own 100% sentence-selection condition, which should reconstruct the full document, gives 94.35–94.78% depending on method, with Ranked Normalized at 94.78% and the 90% row at 94.90%; these values bracket and even exceed the 94.706% baseline. The claimed 0.33% drop is therefore not a controlled comparison to a baseline measured in the same setup; the central efficiency result rests on a self-cited number that the paper's own table does not reproduce.

full rationale

The paper's TF-IDF sentence ranking itself is an empirical heuristic with a non-circular test: sentences are scored by within-article TF-IDF statistics and the resulting subsets are evaluated on held-out labels. If the only limitation were the heuristic's semantic weakness, the circularity score would be low. However, the headline efficiency claim ('only 0.33 percent drop ... reducing inference latency by 43 percent') is not an independent prediction. Section 4.1 reports that λ2 = 0.7 'peaks' on the test-set accuracies in Table 2, and Section 4.2 presents the 40% row of Table 4 (94.39%) as the best efficiency/accuracy operating point after scanning 10–100% coverage; no held-out validation or seed variance is reported, so the best configuration is selected on the same test set used to report the final number. In addition, the full-context baseline 94.706% is imported from prior L3Cube resources (Section 3.1), while Table 4's own 100% condition gives 94.35–94.78% depending on method, with 94.78% for Ranked Normalized and 94.90% at 90%; the baseline is not reproduced in the same pipeline, so the 0.33% drop is not a controlled comparison. These two issues make the central claim partially circular, even though the method comparisons (ranked > first > random > last) retain independent empirical content.

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

The method relies on standard TF-IDF machinery and a single dataset split. The three listed free parameters (lambda_2, selection percentage, and the choice of normalized ranking) are tuned on the test results and directly shape the headline accuracy number. No invented entities are introduced.

free parameters (3)
  • lambda_2 = 0.7
    The weighting factor balancing normalized TF-IDF and sentence length was tuned on the test set; Table 2 shows accuracy varying with lambda_2 and 0.7 chosen as optimal.
  • percentage_of_sentences = 40%
    The headline result uses 40% context with normalized ranking; other percentages yield different accuracies in Table 4, so this value is selected post hoc.
  • selection_mode = ranked_normalized
    The 0.33% drop is reported for the ranked-normalized variant; plain ranked at 40% gives 93.98%, a larger drop. The best-performing variant is chosen from the test results.
assumptions (3)
  • domain assumption TF-IDF sentence scores computed within an article rank sentences by their relevance for classification
    The method relies on this to select informative sentences; Section 3.2 defines the scoring but provides no evidence linking the scores to class-relevant content.
  • domain assumption Fine-tuning MahaBERT-v2 on truncated inputs preserves label information
    The paper assumes that reducing input to selected sentences still allows the model to learn the category; this is the empirical hypothesis being tested.
  • domain assumption The MahaNews LDC test set is representative and fixed
    Single split; no cross-validation or repeated runs are reported.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Improving the Efficiency of Long Document Classification using Sentence Ranking Approach." pith.science (2026). https://pith.science/paper/5AHVH2OH

@misc{pith2026250607248,
  author       = {Pith},
  title        = {Pith review of: Improving the Efficiency of Long Document Classification using Sentence Ranking Approach},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5AHVH2OH}},
  note         = {Machine review of arXiv:2506.07248}
}
read the original abstract

Long document classification poses challenges due to the computational limitations of transformer-based models, particularly BERT, which are constrained by fixed input lengths and quadratic attention complexity. Moreover, using the full document for classification is often redundant, as only a subset of sentences typically carries the necessary information. To address this, we propose a TF-IDF-based sentence ranking method that improves efficiency by selecting the most informative content. Our approach explores fixed-count and percentage-based sentence selection, along with an enhanced scoring strategy combining normalized TF-IDF scores and sentence length. Evaluated on the MahaNews LDC dataset of long Marathi news articles, the method consistently outperforms baselines such as first, last, and random sentence selection. With MahaBERT-v2, we achieve near-identical classification accuracy with just a 0.33 percent drop compared to the full-context baseline, while reducing input size by over 50 percent and inference latency by 43 percent. This demonstrates that significant context reduction is possible without sacrificing performance, making the method practical for real-world long document classification tasks.

Figures

Figures reproduced from arXiv: 2506.07248 by the authors.

Figure 1
Figure 1. Illustration of key idea — selective sentence processing for efficient document classification. The [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Workflow of ranked approach for sentence selection — The diagram illustrates a ranked sentence selection [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Sentence selection approaches — The image illustrates various sentence selection approaches used for [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (3 more)
Figure 5
Figure 5. Figure 5: Normalized sentence-wise accuracy graph — [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: Percentage-wise accuracy graph — The graph [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 8
Figure 8. Figure 8: Evaluation time graph for percentage-wise [PITH_FULL_IMAGE:figures/full_fig_p010_8.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

34 extracted references · 25 canonical work pages

  1. [1]

    URL: " 'urlintro :=

    ENTRY address author booktitle chapter edition editor howpublished institution journal key month note number organization pages publisher school series title type volume year eprint doi pubmed url lastchecked label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block STRINGS urlintro eprinturl eprintpr...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

  3. [3]

    L3cube-indicnews: News-based short text and long document classification datasets in indic languages

    Mirashi Aishwarya, Sonavane Srushti, Lingayat Purva, Padhiyar Tejas, and Joshi Raviraj. L3cube-indicnews: News-based short text and long document classification datasets in indic languages. In Proceedings of the 20th International Conference on Natural Language Processing (ICON), pages 442--449, 2023

    work page 2023

  4. [4]

    Recent advances in long documents classification using deep-learning

    Muhammad Al-Qurishi. Recent advances in long documents classification using deep-learning. In Proceedings of the 5th International Conference on Natural Language and Speech Processing (ICNLSP 2022), 2022. doi:10.18653/v1/2022.icnlsp-1.12

    work page doi:10.18653/v1/2022.icnlsp-1.12 2022

  5. [5]

    New alignment methods for discriminative book summarization, 2013

    David Bamman and Noah Smith. New alignment methods for discriminative book summarization, 2013

    work page 2013

  6. [6]

    Peters, and Arman Cohan

    Iz Beltagy, Matthew E. Peters, and Arman Cohan. Longformer: The long-document transformer, 2020

    2020

  7. [7]

    An improved text sentiment classification model using tf-idf and next word negation, 2020

    Bijoyan Das and Sarit Chakraborty. An improved text sentiment classification model using tf-idf and next word negation, 2020

    work page 2020

  8. [8]

    Selvakumar, and P.J.A

    Mamata Das, K. Selvakumar, and P.J.A. Alphonse. A comparative study on tf-idf feature weighting method and its analysis using unstructured dataset, 2023

    work page 2023

Show all 34 references
  1. [9]

    Devlin, M

    J. Devlin, M. W. Chang, K. Lee, and K. Toutanova. Bert: Pre-training of deep bidirectional transformers for language understanding, 2018

    2018

  2. [10]

    Long document classification from local word glimpses via recurrent attention learning

    Jun He. Long document classification from local word glimpses via recurrent attention learning. IEEE Access, 2019. doi:10.1109/ACCESS.2019.2907992

    2019

  3. [11]

    Indic-transformers: An analysis of transformer language models for indian languages, 2020

    Kushal Jain, Adwait Deshpande, Kumar Shridhar, et al. Indic-transformers: An analysis of transformer language models for indian languages, 2020

    2020

  4. [12]

    L3cube-mahacorpus and mahabert: Marathi monolingual corpus, marathi bert language models, and resources

    Raviraj Joshi. L3cube-mahacorpus and mahabert: Marathi monolingual corpus, marathi bert language models, and resources. In Proceedings of the WILDRE-6 Workshop within the 13th Language Resources and Evaluation Conference, pages 97--101, 2022

    2022

  5. [13]

    Hierarchical neural network approaches for long document classification

    Snehal Ishwar Khandve, Vedangi Kishor Wagh, Apurva Dinesh Wani, Isha Mandar Joshi, and Raviraj Bhuminand Joshi. Hierarchical neural network approaches for long document classification. In Proceedings of the 2022 14th International Conference on Machine Learning and Computing, ...

    2022

  6. [14]

    Research paper classification systems based on tf-idf and lda schemes

    Sang-Woon Kim and Joon-Min Gil. Research paper classification systems based on tf-idf and lda schemes. Human-centric Computing and Information Sciences, 2019. doi:10.1186/s13673-019-0192-7

    2019 doi

  7. [15]

    A novel document classification algorithm based on statistical features and attention mechanism, 2018

    Chao Li, Yanfen Cheng, and Hongxia Wang. A novel document classification algorithm based on statistical features and attention mechanism, 2018

    2018

  8. [16]

    C. z. Liu, Y. x. Sheng, Z. q. Wei, and Y. Q. Yang. Research of text classification based on improved tfi-df algorithm. In 2018 IEEE International Conference of Intelligent Robotic and Control Engineering (IRCE), 2018 a . doi:10.1109/IRCE.2018.8492945

    2018

  9. [17]

    L. Liu, K. Liu, Z. Cong, J. Zhao, Y. Ji, and J. He. Long length document classification by local convolutional feature aggregation. Algorithms, 2018 b . doi:10.3390/a11080109

    2018 doi

  10. [18]

    André F. T. Martins, António Farinhas, Marcos Treviso, et al. Sparse and continuous attention mechanisms, 2020

    2020

  11. [19]

    Minaee, N

    S. Minaee, N. Kalchbrenner, E. Cambria, et al. Deep learning based text classification: A comprehensive review, 2021

    2021

  12. [20]

    L3cube-mahanews: News-based short text and long document classification datasets in marathi

    Saloni Mittal, Vidula Magdum, Sharayu Hiwarkhedkar, Omkar Dhekane, and Raviraj Joshi. L3cube-mahanews: News-based short text and long document classification datasets in marathi. In International Conference on Speech and Language Technologies for Low-resource Languages, pages ...

    2023

  13. [21]

    G. Moro. Efficient memory-enhanced transformer for long-document summarization in low-resource regimes. Sensors, 2023. doi:10.3390/s23073542

    2023 doi

  14. [22]

    Efficient classification of long documents using transformers

    Hyunji Park, Yogarshi Vyas, and Kashif Shah. Efficient classification of long documents using transformers. In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), 2022. doi:10.18653/v1/2022.acl-short.79

    2022 doi

  15. [23]

    Lnlf-bert: Transformer for long document classification with multiple attention levels

    Linh Manh Pham and Hoang Cao The. Lnlf-bert: Transformer for long document classification with multiple attention levels. IEEE Access, 2024. doi:10.1109/ACCESS.2024.3492102

    2024

  16. [24]

    Multi-class text classification using bert-based active learning, 2021

    Sumanth Prabhu, Moosa Mohamed, and Hemant Misra. Multi-class text classification using bert-based active learning, 2021

    2021

  17. [25]

    Topic classification using the long short-term memory (lstm) method with fasttext feature expansion on twitter

    Bella Adriani Putri and Erwin Budi Setiawan. Topic classification using the long short-term memory (lstm) method with fasttext feature expansion on twitter. 2023. doi:10.1109/ICoDSA58501.2023.10277033

    2023

  18. [26]

    Text mining: Use of tf-idf to examine the relevance of words to documents

    Shahzad Qaiser and Ramsha Ali. Text mining: Use of tf-idf to examine the relevance of words to documents. IJCA, 2018. doi:10.5120/ijca2018917395

    2018 doi

  19. [27]

    B. Song. State space models based efficient long documents classification. Journal of Intelligent Learning Systems and Applications, 2024. doi:10.4236/jilsa.2024.163009

    2024

  20. [28]

    How to fine-tune bert for text classification?, 2020

    Chi Sun, Xipeng Qiu, Yige Xu, et al. How to fine-tune bert for text classification?, 2020

    2020

  21. [29]

    Y. Tay, M. Dehghani, S. Abnar, et al. Long range arena: A benchmark for efficient transformers, 2021

    2021

  22. [30]

    A deep neural network approach using convolutional network and long short term memory for text sentiment classification

    Shoryu Teragawa, Lei Wang, and Ruixin Ma. A deep neural network approach using convolutional network and long short term memory for text sentiment classification. In 2021 IEEE 24th International Conference on Computer Supported Cooperative Work in Design (CSCWD), 2021. doi:10....

    2021

  23. [31]

    Comparative study of long document classification

    Vedangi Wagh, Snehal Khandve, Isha Joshi, Apurva Wani, Geetanjali Kale, and Raviraj Joshi. Comparative study of long document classification. In TENCON 2021-2021 IEEE Region 10 Conference (TENCON), pages 732--737. IEEE, 2021

    2021

  24. [32]

    Hierarchical attention networks for document classification

    Zichao Yang, Diyi Yang, Chris Dyer, et al. Hierarchical attention networks for document classification. In Proceedings of the 2016 Conference of the North A merican Chapter of the Association for Computational Linguistics: Human Language Technologies , 2016. doi:10.18653/v1/N16-1174

    2016 doi

  25. [33]

    O. Zafrir. Q8bert: Quantized 8bit bert, 2019

    2019

  26. [34]

    Zaheer, J

    M. Zaheer, J. Ainslie, G. Guruganesh, et al. Big bird: Transformers for longer sequences. In Proceedings of NeurIPS, pages 702--709, 2020. doi:10.48550/arXiv.2007.14062

    2020 doi arXiv

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.