Pith. sign in

REVIEW 5 major objections 6 minor 62 references

Design of intelligent proofreading system for English translation based on CNN and BERT

T0 review · 5 major / 6 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 A hybrid CNN-BERT system claims state-of-the-art English-German translation proofreading, reporting 90% accuracy and 89.37% F1 on WMT and OpenSubtitles.

desk verdict This paper's headline metrics are internally impossible and no protocol is provided, so the central empirical claim fails despite a plausible high-level idea. read the letter →

arxiv 2506.04811 v1 pith:37CWH46G submitted 2025-06-05 cs.CL

Feijun Liu , Huifeng Wang , Kun Wang , Yizhen Wang This is my paper
classification cs.CL
keywords translationproofreadingCNNBERTerrordetectioncorrectionEnglish-Germanalignmentsequencemodeling
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 single hybrid model can automatically proofread machine-translated English by detecting common errors—mistranslations, omissions, and word-order problems—and suggesting corrections that keep the original meaning. The proposed pipeline combines convolutional filters that capture local n-gram patterns with BERT's bidirectional context representations, then feeds flagged errors to a GRU decoder supported by translation memory. The model is trained end-to-end on English-German parallel data from WMT and OpenSubtitles, supervised jointly by detection and correction losses. If the reported results hold, the method would reduce the human post-editing burden that currently limits practical use of machine translation.

What carries the argument

The load-bearing mechanism is the alignment-discrepancy detector followed by a coherence-preserving corrector. A CNN with kernel sizes 1 through 5 produces local n-gram feature maps; BERT's stacked bidirectional self-attention turns the full sequence into context-aware token vectors; a CRF layer labels error spans such as omissions, replacements, insertions, and word-order faults; and a GRU decoder, aided by translation memory, rewrites the flagged spans. The error-detection loss is cross-entropy over predicted versus actual source–target token alignments, while the correction stage is trained to produce edits that stay faithful to the source meaning.

What would settle it

Take a held-out set of English-German machine translations with human-annotated error spans, run the detector, and measure its precision and recall on exactly those annotated spans. If the detector's score on human-labeled omissions and word-order errors falls well below the reported 90% accuracy and 89.37% F1, the alignment-supervision assumption is falsified.

Watch

Extended reading notes

Core claim

On its own terms, the central claim is that the CNN-BERT hybrid is a new state of the art for translation proofreading: it detects errors by comparing source and target token alignments under BERT contextual embeddings, and it corrects them with a GRU-based post-editing model that consults translation memory. The paper reports 90% accuracy, 89.37% F1, and 16.24% MSE on its test setup, and says these numbers exceed five recent proofreading techniques by more than 10% overall. The supporting evidence includes ablations over convolution kernel sizes and batch-normalization sizes, plus a computational-time comparison in which the proposed model is the fastest at 2.12 seconds.

Load-bearing premise

The error detector is supervised only by alignment differences between source and target tokens, so the whole system assumes that misalignment is a reliable proxy for real translation errors; if alignment noise does not match human-annotated errors, the reported detection accuracy will not hold on real post-editing tasks.

Editorial extensions

If this is right

  • If the reported accuracy and F1 hold, automatic post-editing could flag omissions and word-order errors without human-annotated error labels, using alignment differences as supervision.
  • Translation memory plus GRU decoding would let corrections preserve source meaning, which is the main requirement for practical post-editing tools.
  • The hybrid architecture combines local and global context in one end-to-end trainable pipeline, so gains on WMT and OpenSubtitles would extend to other English-German text domains.
  • The claimed 2.12-second runtime makes the model feasible for interactive or real-time proofreading, unlike slower benchmark systems.

Reading between the lines

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

  • A direct test against human-annotated error spans would clarify whether alignment noise really corresponds to translation errors; the paper does not report such a test.
  • Because the architecture is not English-German-specific, the same alignment-supervision recipe could be tried on other language pairs, though low-resource pairs may lack the parallel data needed for the translation-memory component.
  • A natural next benchmark would pit this pipeline against a general-purpose large language model used as a post-editor; that comparison is absent from the paper and would test whether the reported 'over 10%' margin is specific to the five chosen baselines.
  • The alignment-based detector may conflate legitimate reorderings with errors in morphologically rich target languages, so multilingual evaluation would need language-specific alignment constraints.
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 / 6 minor

Summary. The paper proposes a hybrid CNN-BERT sequence-modeling architecture for proofreading English-German machine translation. The method extracts local n-gram features with CNN filters, builds contextual representations with BERT, detects translation errors through token alignment and a CRF-style sequence layer, and proposes corrections with GRU decoders and translation memory. The system is trained end-to-end on WMT and OpenSubtitles parallel corpora, and the abstract claims 90% accuracy, 89.37% F1, and 16.24% MSE, exceeding recent techniques by over 10%. The main body reports ablations over kernel sizes and batch-normalization sizes, comparative tables against five baselines, and computational-time measurements.

Significance. If the empirical results were valid, the paper would describe a practically useful combination of local and contextual models for machine-translation post-editing, with a clearly stated task and an end-to-end training objective. However, the significance assessment is entirely contingent on the credibility of the reported numbers, and that credibility fails: the headline metrics are arithmetically inconsistent, the final model row cannot be traced to any single configuration in the ablation tables, and no reproducible protocol, code, data, or held-out evaluation details are provided. The paper therefore does not currently establish its central claim of state-of-the-art proofreading performance.

major comments (5)
  1. [Table 5, Combined CNN and BERT Model row] The reported F1 score is mathematically impossible given the reported precision and recall. For P=88.34% and R=85.34%, the standard definition F1 = 2PR/(P+R) gives approximately 86.81%, and since F1 is a harmonic mean it must lie between P and R. The reported value 89.37% exceeds both. Even under macro-averaging, macro-F1 cannot exceed the arithmetic mean of macro-P and macro-R, which is 86.84%. This contradicts the abstract's headline F1 claim.
  2. [Tables 2, 3, and 5] The RMSE and MSE columns violate their defining relationship. In Table 5, the proposed model reports RMSE=8.35% and MSE=16.24%, but sqrt(16.24) is approximately 4.03, not 8.35. Similar contradictions appear throughout Tables 2 and 3, e.g., Table 2 BN=16 reports RMSE=15.47 and MSE=2.57, and BN=32 reports RMSE=4.35 and MSE=14.54. Unless the paper gives a nonstandard definition of these metrics, every RMSE/MSE pair in the paper is internally inconsistent, so the error-based performance claims are not interpretable.
  3. [Section 4.5 versus Sections 4.1-4.2] The final 'Combined CNN and BERT Model' row in Table 5 does not correspond to any single configuration reported in the ablation tables. Its RMSE, MSE, and MAE values (8.35, 16.24, 18.65) are identical to the BN=64 row of Table 2, but that row reports precision=65.87%, recall=75.68%, and F1=60.24%, not the 88.34/85.34/89.37 values in Table 5. The 90.00% accuracy in Table 5 also appears as the BN=64 accuracy in Table 2, yet the other metrics in that row are completely different. The headline result is therefore not the evaluation of a fixed model, and the abstract's central empirical claim is unsupported.
  4. [Sections 3.2-3.5 and 4.1-4.5] No training, validation, or test protocol is specified. The paper does not state how data were split, how kernel size and batch-normalization size were selected, whether any held-out test set was used, or how the final model was chosen. The reported optimal values appear to be selected from the same metric tables that are later used to claim superiority. With no out-of-sample evaluation, no error bars, and no statistical significance testing, the comparative claim of 'state-of-the-art' performance and the 'over 10%' improvement is not established.
  5. [Section 3.5.1] The detection module is supervised only by cross-entropy loss on alignment between source and target tokens, as defined in Eq. (1). The manuscript never validates this alignment-based signal against human-annotated translation error labels such as omissions or word-order errors. Since the central claim includes accurate detection of these specific error types, the paper needs an explicit evaluation of detection quality against such annotations; otherwise the high precision/recall/F1 numbers, even if internally consistent, would not demonstrate the claimed proofreading ability.
minor comments (6)
  1. [Section 3.5 and Figure 4] The text alternates between an attention-based alignment detector and a CRF sequence layer, while Figure 4 labels the layer 'CRT' instead of 'CRF'. The manuscript should clarify which detection mechanism is actually used and fix the label.
  2. [Section 1 and References] Many cited works on DC motors, fuzzy control, fault diagnosis, and formation control (e.g., references [4], [5], [12], [15], [16], [18], [24], [25], [45], [46]) are not connected to translation proofreading. These citations appear to be padding and should be removed or properly integrated into the discussion.
  3. [Section 4.5, Table 5] The baseline papers cited for comparison are largely from sentiment analysis and named-entity recognition (e.g., [48], [49], [50], [51]), not from translation proofreading or automatic post-editing. The comparative claim would need baselines evaluated on the same task and data.
  4. [Equations (1)-(5)] Several equations are ambiguous: in Eq. (1) the alignment variable y_ij is not formally defined; in Eq. (3) the denominator is not written as a proper normalization over all possible label sequences; and in Eq. (4) the same symbol h_t denotes both the hidden state and the candidate hidden state. These should be clarified.
  5. [Figures 10 and 11] The scatter plots are described as showing seven feature clusters, but the axes are unlabeled and the plotted coordinates are not tied to any actual feature values, so the figures cannot be interpreted.
  6. [Section 4.5] The paper repeatedly mentions BLEU as an evaluation metric, but no BLEU scores are reported in any table or figure. Add the BLEU results or remove the claim.

Circularity Check

2 steps flagged · score 6.0 of 10

Headline metrics are selected hyperparameter optima presented as a benchmark, and the error-detection objective is self-defined as alignment prediction.

  1. self definitional [Section 3.5 and Eq. (1) (3.5.1)]
    "Based on the alignments, it flags translation errors like word omissions, replacements, insertions as well as incorrect word order. ... 𝐿𝑐𝑒 = − ∑ ∑ 𝑦𝑖𝑗 log (𝑦^𝑖𝑗) ... 𝑦𝑖𝑗 denotes the alignment that actually occurs (that is, whether a source token aligns with a target token), whereas 𝑦^𝑖𝑗 reflects the alignment probability that the model predicts."

    The supervision for the detection module is the alignment matrix y_ij, and the model output yhat_ij is the predicted alignment; the paper's 'translation errors' (omissions, replacements, insertions, word order) are identified from those same alignments. No independent human-annotated error labels are used, so 'detecting translation errors' reduces by construction to reproducing the alignment supervision. The detector's reported success is therefore a measure of alignment prediction, not of translation-error detection.

  2. fitted input called prediction [Section 4, Tables 1-3 and Table 5; Abstract]
    "Experiments attain a 90% accuracy, 89.37% F1, and 16.24% MSE, exceeding recent proofreading techniques by over 10% overall. ... Combined CNN and BERT Model 88.34 85.34 89.37 90.00 8.35 16.24 18.65"

    The abstract's headline numbers are the same quantities produced by the hyperparameter sweeps in Tables 1-3: accuracy 90.00 appears for kernel size 3 (Table 1) and for BN=64 (Table 2), and MSE 16.24 appears for BN=64 (Table 2). The paper tunes kernel and batch-normalization sizes against precision/recall/F1/accuracy/RMSE/MSE and then reports the selected values as the final 'Combined CNN and BERT Model' comparison row. The claim of >10% superiority over recent systems is thus a selected optimum reported as a benchmark, rather than an out-of-sample prediction; the row itself is not even internally consistent (F1=89.37 exceeds the harmonic mean of P=88.34 and R=85.34).

full rationale

The central detection step is circular in the self-definitional sense: 'translation errors' are read off the alignment matrix that Eq. (1) is trained to predict, with no external error annotation, so detection accuracy cannot be distinguished from alignment prediction. Independently, the state-of-the-art claim is a fitted-input-called-prediction issue: the final comparison row reuses accuracy and MSE values from the kernel and batch-size sweeps, so the reported >10% advantage is an in-sample selected optimum rather than a derived prediction. The paper does not rely on load-bearing self-citations; apparent co-author references such as [16]/[52] are unrelated to the proofreading architecture. I am not counting the metric inconsistencies (F1 greater than the harmonic mean of precision and recall, RMSE not equal to the square root of MSE, and duplicated or contradictory table values) as circularity; those are correctness and verification issues rather than reductions to inputs. No code, data, or test protocol is supplied, so the empirical claims are also unverifiable. The score reflects the two reduction-by-construction problems, not the additional correctness problems.

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

The paper's performance claim depends on two explicitly tuned hyperparameters (kernel size and batch size), an unstated assumption that alignment differences can serve as error labels, and an unvalidated assumption that selecting configurations on the evaluation metrics generalizes. No code or data are provided.

free parameters (3)
  • Convolution kernel size = 3 (best accuracy 90.00% in Table 1)
    Chosen by comparing kernel sizes 1, 3, 4, and 5 against precision, recall, F1, and accuracy; the best-accuracy configuration is used in the final model.
  • Batch-normalization size = 64 (best accuracy 90.00% on WMT in Table 2)
    Chosen by comparing sizes 16, 32, 64, 128, and 256 on the WMT dataset; the best-accuracy configuration is used in the final comparison.
  • Loss weights for detection and correction losses = Not stated
    Section 3.2 says 'multiple loss functions supervise detection and correction' but the relative weights are never reported, so the final metric trade-off depends on unreported choices.
assumptions (3)
  • standard math Standard CNN, BERT, CRF, GRU, and BLEU formulas (Eqs. 1-5) apply unchanged to the proofreading task.
    The paper invokes these textbook equations in Section 3.5 without modification or derivation.
  • domain assumption Parallel English-German corpora provide implicit supervision for error detection through alignment differences without explicit error labels.
    Section 3.2 and 3.5 train detection with cross-entropy over alignments but never describe how gold error labels are constructed.
  • ad hoc to paper Selecting kernel size and batch size on the evaluation metrics yields a generalizable model.
    Tables 1 and 2 choose configurations that maximize accuracy, and the same selected model is then reported as state-of-the-art in Table 5.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Design of intelligent proofreading system for English translation based on CNN and BERT." pith.science (2026). https://pith.science/paper/37CWH46G

@misc{pith2026250604811,
  author       = {Pith},
  title        = {Pith review of: Design of intelligent proofreading system for English translation based on CNN and BERT},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/37CWH46G}},
  note         = {Machine review of arXiv:2506.04811}
}
read the original abstract

Since automatic translations can contain errors that require substantial human post-editing, machine translation proofreading is essential for improving quality. This paper proposes a novel hybrid approach for robust proofreading that combines convolutional neural networks (CNN) with Bidirectional Encoder Representations from Transformers (BERT). In order to extract semantic information from phrases and expressions, CNN uses a variety of convolution kernel filters to capture local n-gram patterns. In the meanwhile, BERT creates context-rich representations of whole sequences by utilizing stacked bidirectional transformer encoders. Using BERT's attention processes, the integrated error detection component relates tokens to spot translation irregularities including word order problems and omissions. The correction module then uses parallel English-German alignment and GRU decoder models in conjunction with translation memory to propose logical modifications that maintain original meaning. A unified end-to-end training process optimized for post-editing performance is applied to the whole pipeline. The multi-domain collection of WMT and the conversational dialogues of Open-Subtitles are two of the English-German parallel corpora used to train the model. Multiple loss functions supervise detection and correction capabilities. Experiments attain a 90% accuracy, 89.37% F1, and 16.24% MSE, exceeding recent proofreading techniques by over 10% overall. Comparative benchmarking demonstrates state-of-the-art performance in identifying and coherently rectifying mistranslations and omissions.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

62 extracted references · 52 canonical work pages

  1. [1]

    and Huang, Y., 2023

    Li, S. and Huang, Y., 2023. BAS -ALSTM: analyzing the efficiency of artificial intelligence-based English translation system. Journal of Ambient Intelligence and Humanized Computing, pp.1-13

    work page 2023

  2. [2]

    Automatic error calibration system for English semantic translation based on machine learning

    Wei, Z., 2023. Automatic error calibration system for English semantic translation based on machine learning. International Journal of Industrial and Systems Engineering , 43(3), pp.301-316

    work page 2023

  3. [3]

    and Shao, D., 2023

    Ma, R. and Shao, D., 2023. English Translation proofreading System based on Information Technology: Construction of semantic Ontology Translation Model. Applied Artificial Intelligence, 37(1), p.2201145

    work page 2023

  4. [4]

    PID tracking control design for electrical DC motors with experimental validation H 2 norm

    Aslam, Muhammad Shamrooz, Summera Shamrooz, Hazrat Bilal, and Asim Aziz. "PID tracking control design for electrical DC motors with experimental validation H 2 norm." Journal of Control and Decision (2024): 1-15, https://doi.org/10.1080/23307706.2024.2414378

    work page arXiv 2024

  5. [5]

    Enhancing Trajectory Tracking and Vibration Control of Flexible Robots With Hybrid Fuzzy ADRC and Input Shaping,

    Y. Tian, A. Yahya et al. , "Enhancing Trajectory Tracking and Vibration Control of Flexible Robots With Hybrid Fuzzy ADRC and Input Shaping," in IEEE Access, vol. 12, pp. 150574-150591, 2024, doi: 10.1109/ACCESS.2024.3453944

    work page arXiv 2024

  6. [6]

    and Jia, C., 2021, December

    Cairang, H., Jia, S. and Jia, C., 2021, December. Research on error correction method of Tibetan text based on deep learning. In Second IYSF Academic Symposium on Artificial Intelligence and Computer Engineering (Vol. 12079, pp. 325-335). SPIE

    work page 2021

  7. [7]

    and Mayhew, S., 2021

    Adelani, D.I., Abbott, J., Neubig, G., D’souza, D., Kreutzer, J., Lignos, C., Palen -Michel, C., Buzaaba, H., Rijhwani, S., Ruder, S. and Mayhew, S., 2021. MasakhaNER: Named entity recognition for African languages. Transactions of the Association for Computational Linguistics, 9, pp.1116-1131

    work page 2021

  8. [8]

    Modeling of Asynchronous Mode -Dependent Delays in Stochastic Markovian Jumping Modes Based on Static Neural Networks for Robotic Manipulators,

    Sumeera. S., et al., "Modeling of Asynchronous Mode -Dependent Delays in Stochastic Markovian Jumping Modes Based on Static Neural Networks for Robotic Manipulators," in IEEE Transactions on Automation Science and Engineering, vol. 22, pp. 13398-13410, 2025, doi: 10.1109/TASE.2025.3552645

    arXiv 2025

Show all 62 references
  1. [9]

    DSQN: Robust path planning of mobile robot based on deep spiking Q-network

    Kumar, Aakash, et al. "DSQN: Robust path planning of mobile robot based on deep spiking Q-network." Neurocomputing 634 (2025): 129916 , https://doi.org/10.1016/j.neucom.2025.129916

    2025

  2. [10]

    and Gu, N., 2023

    Abbas, R. and Gu, N., 2023. Improving deep learning -based image super-resolution with residual learning and perceptual loss using SRGAN model. Soft Computing, pp.1-17

    2023

  3. [11]

    and Rauf, H.T., 2023

    Alawadh, H.M., Alabrah, A., Meraj, T. and Rauf, H.T., 2023. Attention -Enriched Mini- BERT Fake News Analyzer Using the Arabic Language. Future Internet, 15(2), p.44

    2023

  4. [12]

    Self-Triggered Scheme Design for Takagi-Sugeno Fuzzy Model Based on Mismatch Premise Variable With Time-Varying Delay,

    Muhammad. S. A., et al., "Self-Triggered Scheme Design for Takagi-Sugeno Fuzzy Model Based on Mismatch Premise Variable With Time-Varying Delay," in IEEE Transactions on Automation Science and Engineering, vol. 22, pp. 15536-15548, 2025, doi: 10.1109/TASE.2025.3570089

    2025

  5. [13]

    A Consumer Electronics-Enhanced UAV System for Agricultural Farm Tracking With Fuzzy SMO and Actuator Fault Detection Control Algorithms,

    B. Hazrat, et al., "A Consumer Electronics-Enhanced UAV System for Agricultural Farm Tracking With Fuzzy SMO and Actuator Fault Detection Control Algorithms," in IEEE Transactions on Consumer Electronics, doi: 10.1109/TCE.2025.3563993

  6. [14]

    Analysis of Chinese Text Automatic Proofreading Technology

    Zhang, Y., 2022. Analysis of Chinese Text Automatic Proofreading Technology. International Journal of Education and Humanities, 4(1), pp.169-172

    2022

  7. [15]

    H∞ Delayed Filtering of Markov Jump Fuzzy Systems in Consumer Electronics: Input –Output Analysis,

    M. S. Aslam, et al. , "H∞ Delayed Filtering of Markov Jump Fuzzy Systems in Consumer Electronics: Input –Output Analysis," in IEEE Transactions on Consumer Electronics , doi: 10.1109/TCE.2025.3565105

    2025

  8. [17]

    and Choi, W.J., 2023

    Lee, S.W. and Choi, W.J., 2023. Utilizing ChatGPT in clinical research related to anesthesiology: a comprehensive review of opportunities and limitations. Anesthesia and Pain Medicine, 18(3), pp.244-251

    2023

  9. [18]

    Chang, Wj. et al. Indirect adaptive observer control (I-AOC) design for truck–trailer model based on T –S fuzzy system with unknown nonlinear function. Complex Intell. Syst. 10, 7311 –7331 (2024). https://doi.org/10.1007/s40747-024-01544-7

    2024 doi

  10. [19]

    78, 101020, 2024

    Shamrooz S, et al., Fuzzy PD-sliding mode control design for networked system with time delays, European Journal of Control, vol. 78, 101020, 2024. https://doi.org/10.1016/j.ejcon.2024.101020

    2024

  11. [20]

    and Zaki, T., 2023

    Bensoltane, R. and Zaki, T., 2023. Combining BERT with TCN -BiGRU for enhancing Arabic aspect category detection. Journal of Intelligent & Fuzzy Systems, (Preprint), pp.1- 14

    2023

  12. [21]

    and Baz, M., 2022

    Prottasha, N.J., Sami, A.A., Kowsher, M., Murad, S.A., Bairagi, A.K., Masud, M. and Baz, M., 2022. Transfer learning for sentiment analysis using BERT based supervised fine - tuning. Sensors, 22(11), p.4157

    2022

  13. [22]

    and Zhang, Q., 2022

    Duan, K., Du, S., Zhang, Y., Lin, Y., Wu, H. and Zhang, Q., 2022. Enhancement of Question Answering System Accuracy via Transfer Learning and BERT. Applied Sciences, 12(22), p.11522

    2022

  14. [23]

    and Bos, J., 2020

    Van Noord, R., Toral, A. and Bos, J., 2020. Character-level representations improve DRS- based semantic parsing Even in the age of BERT. arXiv preprint arXiv:2011.04308

    2020 arXiv

  15. [24]

    Kumar, A., Wang, S., Shaikh, A.M. et al. Building on prior lightweight CNN model combined with LSTM-AM framework to guide fault detection in fixed -wing UAVs. Int. J. Mach. Learn. & Cyber. 15, 4175–4191 (2024). https://doi.org/10.1007/s13042-024-02141-3

    2024 doi

  16. [25]

    Online Fault Diagnosis of Industrial Robot Using IoRT and Hybrid Deep Learning Techniques: An Experimental Approach,

    H. Bilal, M. S. Obaidat, M. Shamrooz Aslam, J. Zhang, B. Yin and K. Mahmood, "Online Fault Diagnosis of Industrial Robot Using IoRT and Hybrid Deep Learning Techniques: An Experimental Approach," in IEEE Internet of Things Journal, vol. 11, no. 19, pp. 31422-31437, 1 Oct.1, 20...

    2024

  17. [26]

    and Mavridis, I., 2022

    Tsinganos, N., Fouliras, P. and Mavridis, I., 2022. Applying BERT for Early -Stage Recognition of Persistence in Chat -Based Social Engineering Attacks. Applied Sciences, 12(23), p.12353

    2022

  18. [27]

    and Crespi, N., 2020

    Mozafari, M., Farahbakhsh, R. and Crespi, N., 2020. Hate speech detection and racial bias mitigation in social media based on BERT model. PloS one, 15(8), p.e0237861

    2020

  19. [28]

    A blockchain -enabled approach for privacy-protected data sharing in internet of robotic things networks

    Farooq Ahmad, et al. "A blockchain -enabled approach for privacy-protected data sharing in internet of robotic things networks." Human-centric Computing and Information Sciences 14 (2024), https://doi.org/10.22967/HCIS.2024.14.071

    2024 doi

  20. [29]

    Dou, H., Liu, Y., Chen, S. et al. A hybrid CEEMD-GMM scheme for enhancing the detection of traffic flow on highways. Soft Comput 27, 16373–16388 (2023). https://doi.org/10.1007/s00500- 023-09164-y

    2023 doi

  21. [30]

    and Hamid, M.A., 2022

    Keya, A.J., Wadud, M.A.H., Mridha, M.F., Alatiyyah, M. and Hamid, M.A., 2022. AugFake-BERT: handling imbalance through augmentation of fake news using BERT to enhance the performance of fake news classification. Applied Sciences, 12(17), p.8398

    2022

  22. [31]

    Experimental validation of fuzzy PID control of flexible joint system in presence of uncertainties,

    W. Yao, Y. Guo, Y. Wu and J. Guo, "Experimental validation of fuzzy PID control of flexible joint system in presence of uncertainties," 2017 36th Chinese Control Conference (CCC), Dalian, China, 2017, pp. 4192-4197, doi: 10.23919/ChiCC.2017.8028015

    2017

  23. [32]

    Real -Time Lane Detection and Tracking for Advanced Driver Assistance Systems,

    B. Yin, J. Khan, L. Wang, J. Zhang and A. Kumar, "Real -Time Lane Detection and Tracking for Advanced Driver Assistance Systems," 2019 Chinese Control Conference (CCC) , Guangzhou, China, 2019, pp. 6772-6777, doi: 10.23919/ChiCC.2019.8866334

    2019

  24. [33]

    Identification and diagnosis of chronic heart disease: A deep learning -based hybrid approach

    Muhammad Y, Ullah I, Garg S, Choi BJ, Hassan MM. Identification and diagnosis of chronic heart disease: A deep learning -based hybrid approach. Alexandria Engineering Journal. 2025 Jun 1;124:470-83, https://doi.org/10.1016/j.aej.2025.03.025

    2025 doi

  25. [34]

    and Chang, H.T., 2022

    Khan, L., Amjad, A., Ashraf, N. and Chang, H.T., 2022. Multi-class sentiment analysis of urdu text using multilingual BERT. Scientific Reports, 12(1), p.5436

    2022

  26. [35]

    Intelligent English Translation Model Based on Improved GLR Algorithm

    Yang, S., 2023. Intelligent English Translation Model Based on Improved GLR Algorithm. Procedia Computer Science, 228, pp.533-542

    2023

  27. [36]

    An intelligent scoring system for English writing based on artificial intelligence and machine learning

    Huang, Z., 2023. An intelligent scoring system for English writing based on artificial intelligence and machine learning. International Journal of System Assurance Engineering and Management, pp.1-8

    2023

  28. [37]

    and Mwendia, S.N., 2023

    Mpia, H.N., Mburu, L.W. and Mwendia, S.N., 2023. CoBERT: A Contextual BERT model for recommending employability profiles of information technology students in unstable developing countries. Engineering Applications of Artificial Intelligence, 125, p.106728

    2023

  29. [38]

    and Ji, D., 2023

    Zhou, J., Li, F., Teng, C., Liu, Y., Xiang, C. and Ji, D., 2023. MOIT: A Novel task for mining opinions towards implicit targets. Engineering Applications of Artificial Intelligence, 126, p.106841

    2023

  30. [39]

    and Tang, P., 2023

    Haq, I., Qiu, W., Guo, J. and Tang, P., 2023. Correction of whitespace and word segmentation in noisy Pashto text using CRF. Speech Communication, 153, p.102970

    2023

  31. [40]

    Selling research in RA discussion sections through English and Spanish: An intercultural rhetoric approach

    Moreno, A.I., 2021. Selling research in RA discussion sections through English and Spanish: An intercultural rhetoric approach. English for specific purposes, 63, pp.1-17

    2021

  32. [41]

    and Gao, L., 2023

    Shi, C., Xiang, Y., Yu, J., Sood, K. and Gao, L., 2023. Machine translation -based fine- grained comments generation for solidity smart contracts. Information and Software Technology, 153, p.107065

    2023

  33. [42]

    and Yamanaka, T., 2023

    Sugiyama, K. and Yamanaka, T., 2023. Proposals and Methods for Foreign Language Learning Using Machine Translation and Large Language Model. Procedia Computer Science, 225, pp.4750-4757

    2023

  34. [43]

    Automatic proofreading method for English translation accuracy of nano vocabulary

    Liu, Y., 2021. Automatic proofreading method for English translation accuracy of nano vocabulary. Nanotechnology for Environmental Engineering, 6, pp.1-7

    2021

  35. [44]

    Development of Computer Intelligent Proofreading System from the Perspective of English Translation Application

    Zhao, J., 2022, January. Development of Computer Intelligent Proofreading System from the Perspective of English Translation Application. In International Conference on Innovative Computing (pp. 996-1001). Singapore: Springer Nature Singapore

    2022

  36. [45]

    A., & Ullah, I

    Tian, Y., Ali, A., Muhammad, Y., Yahya, A., Izneid, B. A., & Ullah, I. (2024). An Intelligent Approach for Early and Accurate Predication of Cardiac Disease Using Hybrid Artificial Intelligence Techniques. Bioengineering, 11(12), 1290. https://doi.org/10.3390/bioengineering11121290

    2024 doi

  37. [46]

    Trajectory Tracking Control for Disturbed Euler -Lagrange Systems: A Fixed-Time Adaptive Approach

    Anjum, Zeeshan, et al. "Trajectory Tracking Control for Disturbed Euler -Lagrange Systems: A Fixed-Time Adaptive Approach." Journal of Control Engineering and Applied Informatics 26.4 (2024): 15-26, doi: 10.61416/ceai.v26i4.9118

    2024 doi

  38. [47]

    Translation correction of English phrases based on optimized GLR algorithm

    Wang, X., 2021. Translation correction of English phrases based on optimized GLR algorithm. Journal of Intelligent Systems, 30(1), pp.868-880

    2021

  39. [48]

    and Raman, B., 2023

    Kumar, P., Pathania, K. and Raman, B., 2023. Zero -shot learning based cross -lingual sentiment analysis for sanskrit text with insufficient labeled data. Applied Intelligence, 53(9), pp.10096-10113

    2023

  40. [49]

    and Sharma, A., 2023

    Alyoubi, K.H. and Sharma, A., 2023. A Deep CRNN -Based Sentiment Analysis System with Hybrid BERT Embedding. International Journal of Pattern Recognition and Artificial Intelligence, 37(05), p.2352006

    2023

  41. [50]

    Design of Intelligent Recognition English Translation Model Based on Improved Machine Translation Algorithm

    Deng, T., 2022. Design of Intelligent Recognition English Translation Model Based on Improved Machine Translation Algorithm. In 3D Imaging —Multidimensional Signal Processing and Deep Learning: 3D Images, Graphics and Information Technologies, Volume 1 (pp. 233-244). Singapore...

    2022

  42. [51]

    and Ji, X., 2022

    Zhang, W., Wu, Y., Zhang, W., Zhang, Y. and Ji, X., 2022. Named Entity Recognition of Ancient Wine Texts Based on Deep Learning Models. Academic Journal of Science and Technology, 4(2), pp.97-103

    2022

  43. [52]

    Wu, Q., Li, X., Wang, K. et al. Regional feature fusion for on -road detection of objects using camera and 3D-LiDAR in high-speed autonomous vehicles. Soft Comput 27, 18195– 18213 (2023). https://doi.org/10.1007/s00500-023-09278-3

    2023 doi

  44. [53]

    Dou, H., Liu, Y., Chen, S. et al. A hybrid CEEMD -GMM scheme for enhancing the detection of traffic flow on highways. Soft Comput 27, 16373 –16388 (2023). https://doi.org/10.1007/s00500-023-09164-y

    2023 doi

  45. [54]

    Ali, M., Yin, B., Bilal, H. et al. Advanced efficient strategy for detection of dark objects based on spiking network with multi -box detection , Multimedia Tools and Applications, (2023). https://doi.org/10.1007/s11042-023-16852-2

    2023 doi

  46. [55]

    Robust Maximum Power Point Tracking in PV Generation System: A Hybrid ANN-Backstepping Approach With PSO-GA Optimization,

    U. Hussan et al., "Robust Maximum Power Point Tracking in PV Generation System: A Hybrid ANN-Backstepping Approach With PSO-GA Optimization," in IEEE Transactions on Consumer Electronics, doi: 10.1109/TCE.2025.3569871

    2025

  47. [56]

    Gain-scheduled filter design with different time delays for TS fuzzy systems

    Aslam, Muhammad Shamrooz, et al., "Gain-scheduled filter design with different time delays for TS fuzzy systems." Transactions of the Institute of Measurement and Control (2024): 01423312241274009, https://doi.org/10.1177/01423312241274009

    2024 doi

  48. [57]

    Rohra, A., Yin, B. , et al. MSFFNet: multi -scale feature fusion network with semantic optimization for crowd counting. Pattern Anal Applic 28, 21 (2025). https://doi.org/10.1007/s10044-024-01385-7

    2025 doi

  49. [58]

    and Ghasemi, P., 2024

    M.S., Bilal, H., Band, S.S. and Ghasemi, P., 2024. Modeling of nonlinear supply chain management with lead -times based on Takagi -Sugeno fuzzy control model. Engineering applications of artificial intelligence , 133, p.108131 , https://doi.org/10.1016/j.engappai.2024.108131

    2024

  50. [59]

    Kumar, A., Shaikh, A.M., Li, Y. et al. Pruning filters with L1-norm and capped L1-norm for CNN compression. Appl Intell 51, 1152–1160 (2021). https://doi.org/10.1007/s10489- 020-01894-y

    2021 doi

  51. [60]

    Xu, H., Sun, Z., Cao, Y. et al. A data-driven approach for intrusion and anomaly detection using automated machine learning for the Internet of Things. Soft Comput 27, 14469 – 14481 (2023). https://doi.org/10.1007/s00500-023-09037-4

    2023 doi

  52. [61]

    Formation Control of Heterogeneous Multi -Agent Systems Under Fixed and Switching Hierarchies,

    M. Shamrooz Aslam et al ., "Formation Control of Heterogeneous Multi -Agent Systems Under Fixed and Switching Hierarchies," in IEEE Access , vol. 12, pp. 97868 -97882, 2024, doi: 10.1109/ACCESS.2024.3419815

    2024

  53. [62]

    Modeling a Takagi -Sugeno (TS) fuzzy for unmanned aircraft vehicle using fuzzy controller

    Aslam MS, Bi lal H. Modeling a Takagi -Sugeno (TS) fuzzy for unmanned aircraft vehicle using fuzzy controller. Ain Shams Engineering Journal, vol. 15, no. 10, 102984, 2024 , https://doi.org/10.1016/j.asej.2024.102984

    2024

  54. [63]

    Hybrid TrafficAI: A Generative AI Framework for Real -Time Traffic Simulation and Adaptive Behavior Modeling,

    H. Bilal et al., "Hybrid TrafficAI: A Generative AI Framework for Real -Time Traffic Simulation and Adaptive Behavior Modeling," in IEEE Transactions on Intelligent Transportation Systems , doi: 10.1109/TITS.2025.3571041

    2025

Pith tools

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