Integrative Analysis of Financial Market Sentiment Using CNN and GRU for Risk Prediction and Alert Systems

Hongye Zheng; Jinxin Hu; Mengfang Sun; Yingbin Liang; You Wu; Zhenghao Lin

REVIEW 5 major objections 5 minor 2 cited by

Integrative Analysis of Financial Market Sentiment Using CNN and GRU for Risk Prediction and Alert Systems

T0 review · 5 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read The paper argues that a CNN-GRU hybrid network predicts financial market sentiment and risk more accurately than either network alone, with training-set precision of 84.32% and an almost 20% claimed improvement over conventional means.

desk verdict Standard CNN-GRU hybrid whose only reported evidence is training-set accuracy; the central predictive claim is unsubstantiated. read the letter →

arxiv 2412.10199 v1 pith:4HVRDGEN submitted 2024-12-13 cs.LG q-fin.CP

You Wu , Mengfang Sun , Hongye Zheng , Jinxin Hu , Yingbin Liang , Zhenghao Lin This is my paper · ORCID

classification cs.LGq-fin.CP

keywords CNNGRUEmotionAnalysisRiskWarningDeepLearningMarketSentimentS&P500

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 hybrid deep-learning pipeline—convolutional neural network (CNN) for pulling sentiment-bearing phrases out of social media and financial news text, followed by a gated recurrent unit (GRU) for tracking how that sentiment evolves over time—predicts market sentiment and risk more accurately than either network used alone. On a five-year S&P 500 dataset, the authors report that the fused model reaches 84.32% precision, 86.39% recall, and an F1 score of 0.87, against roughly 63% precision for the standalone CNN and GRU, and they characterize the result as an almost 20% improvement in precision over conventional approaches. The claimed payoff is an early-warning system that alerts investors to sentiment turning points and irregular market shocks before they fully show up in price. If the claim holds, the architecture would give financial risk monitoring a practical way to turn unstructured text into an actionable signal.

What carries the argument

The load-bearing mechanism is the CNN-GRU stack, in which a convolutional layer converts tokenized text from social media and finance portals into high-dimensional vector representations by finding local patterns such as sentiment-laden terms and trending themes, and then feeds that feature sequence into a GRU. The GRU's update gate $z_t = \sigma(W_z \cdot [h_{t-1}, x_t])$ and reset gate $r_t = \sigma(W_r \cdot [h_{t-1}, x_t])$ decide how much of the past sentiment state to retain and how much new input to blend in, with candidate state $\tilde{h}_t = \tanh(W \cdot [r_t * h_{t-1}, x_t])$ and final update $h_t = (1 - z_t) * h_{t-1} + z_t * \tilde{h}_t$. This gating is what lets the model hold onto long-range dependencies and avoid the vanishing-gradient problem of plain RNNs, which is the property the paper relies on to turn extracted sentiment features into predictions of future market conditions and risks.

What would settle it

A decisive check is to train the same CNN-GRU on the paper's five-year S&P 500 text-and-price dataset with the sentiment labels randomly permuted relative to the timestamps while keeping prices unchanged; if precision remains near 84.32%, the sentiment channel is not carrying the predictive signal. A second, simpler check is to hold out the final year of the data, train only on the first four years, and report precision, recall, and F1 on the held-out year, since the paper's only reported comparison is on the training set and a test-set number near the 63% baseline would refute the claimed near-20% gain.

Watch

Extended reading notes

Core claim

The central claim is that fusing CNN and GRU is not just a linear succession of two networks but a division of labor: CNN extracts high-dimensional semantic features from text, and GRU learns their temporal progression, so the hybrid can both interpret the current market mood and anticipate its future trajectory and associated risks. In the paper's experiments, the CNN-GRU model predicts equity values on an S&P 500 corpus and, on the training set, outperforms the CNN-only and GRU-only baselines by a wide margin: 84.32% precision and 0.87 F1 versus 62.59% and 0.67 for CNN and 63.17% and 0.69 for GRU. The authors further assert that the model excels at pinpointing sentiment inflection points, such as shifts from bullish to bearish, and at detecting irregular market shocks from unexpected events, crediting the architecture with an almost 20% enhancement in precision over conventional means. An attention-guided mechanism is described as refining the model's focus on the most predictive text segments, supporting the paper's framing of the system as a precise, timely risk-alert tool.

Load-bearing premise

The load-bearing premise is that the sentiment labels assigned to social-media and financial-news text are predictive of future S&P 500 equity values; the paper synchronizes those labels with price histories and trains directly on the pairing without ever testing whether the text-to-price link exists, so if that link is absent the reported accuracy gain would not translate into useful risk alerts.

Editorial extensions

If this is right

If the reported accuracy holds on held-out data, the same CNN-GRU structure can be applied to other unstructured financial text streams such as earnings calls, central bank statements, and regulatory filings to generate earlier risk signals than price-based indicators alone.
The claimed 84.32% training-set precision suggests the hybrid may be suitable for real-time sentiment monitoring, since GRU's smaller parameter count relative to LSTM lowers the computational cost of continuous inference.
The model's stated ability to identify sentiment inflection points implies an alert system that can flag shifts in market mood before they are fully reflected in equity valuations, giving traders a text-driven leading indicator.
The authors suggest the framework can be adapted to multimedia inputs like imagery and audio, so a correct result would extend sentiment risk monitoring beyond text.
A direct corollary of the claim is that the same margin should be reproducible on a held-out test set built from the same five-year corpus before the system is used operationally.

Reading between the lines

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

The paper never tests whether sentiment categories actually predict future S&P 500 values; a fair extension would ablate the text channel by comparing the CNN-GRU against the same network trained on price and volume data alone, and against a version whose sentiment labels are randomly shuffled in time, which would isolate whether the claimed gain comes from sentiment or from the price history itse
The reported metrics come from the training set, so the almost 20% precision claim would be far more convincing if re-evaluated on a chronological hold-out set across different market regimes; this is an editorial check, not the paper's own protocol.
If the sentiment-price link is real, the natural operational step is to turn the model's continuous sentiment scores into discrete alert thresholds such as risk-on or risk-off, with false-alarm rates measured in dollars of avoided drawdown; the paper leaves this translation unspecified.
The attention mechanism is mentioned in the introduction but never compared against a no-attention variant in the experiments, so an extension that quantifies the attention layer's marginal contribution would clarify which part of the architecture is actually responsible for the reported improvement.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

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

Colleague,

This is a standard CNN-GRU hybrid applied to financial sentiment, and the paper's only reported evidence is training-set accuracy. The conclusion's "almost 20% enhancement" is not backed by any test-set number. In its current form, the central claim that the hybrid beats standalone CNN and GRU is unsubstantiated.

What is actually here: The paper lays out the CNN and GRU equations correctly, describes a reasonable preprocessing pipeline that uses a linked data model for sentiment categorization, and synchronizes social media/news text with S&P 500 price data over five years. The writing is clear enough that a reader can see what was attempted. That is the extent of the credit I'd give.

Novelty is thin. The paper's own references include CNN-GRU hybrids for stock forecasting ([17], [23]) and CNN+LSTM for equities ([1]); the architecture is the same stack. The attention mechanism mentioned in the introduction and conclusion is never implemented, designed, or tested. So the only concrete content is an application, and that application lacks real evidence.

The soft spots are not minor. Table I reports precision/recall/F1 on the training set, explicitly. Section IV.C says the data is split into training/validation/test, but no validation or test numbers appear anywhere. The only figure (Fig. 5) doesn't say which split it plots. With a sliding window of size 20 and stride 3, overlapping windows create temporal leakage, so even the training-set figures may be optimistic. There are no error bars, no dataset statistics, and no comparison to the LSTM/ARIMA baselines the text mentions. The sentiment-market premise is also untested: the paper synchronizes sentiment labels with market metrics and proceeds without checking whether the sentiment signal has any predictive content. So the entire risk-prediction claim rests on an unvalidated link.

Who is this for? Possibly a practitioner wanting a rough template for a CNN-GRU sentiment pipeline. As a research contribution, it is not ready for serious refereeing. The missing held-out evaluation is a load-bearing gap, not a stylistic issue. I would tell the authors to add test-set results, proper baselines, and an actual attention mechanism before resubmitting anywhere.

Recommendation: reject, and don't send to review unless the authors provide genuine out-of-sample evidence.

Referee Report

5 major / 5 minor

Summary. The paper proposes a hybrid CNN-GRU model for financial market sentiment analysis and risk prediction, claiming that CNN extracts local textual features and GRU models their temporal evolution, yielding timely risk alerts. The experimental section reports precision, recall, and F1 scores for CNN, GRU, and CNN+GRU in Table I, and the conclusion claims an 'almost 20% enhancement in precision' over conventional methods. The central problem is that the only quantitative comparison is computed on the training set, the promised baselines and attention mechanism are absent from the experiments, and the predictive link between sentiment labels and future market values is never validated.

Significance. If properly validated, a CNN-GRU sentiment model for equity risk prediction would be a plausible and useful contribution to financial machine learning. The paper identifies a relevant problem and describes a standard architecture at a high level, and it states concrete training hyperparameters. However, the current manuscript provides no valid quantitative evidence for the central claim: Table I is explicitly labeled as training-set performance, no held-out metrics are reported, no comparison is made to LSTM, ARIMA, or the other baselines mentioned in the text, and the claimed 20% precision improvement cannot be traced to any table or figure. The work also lacks dataset statistics, a precise description of the sentiment-labeling model, and any reproducibility artifacts such as code or data splits.

major comments (5)

[Section IV.D, Table I] The only quantitative evidence for the superiority of CNN+GRU is Table I, whose caption states 'ON THE TRAINING SET.' Section IV.C says the dataset is split into training, validation, and test sets, but Section IV.D reports no validation or test performance. Training-set metrics measure fit to seen data, not predictive ability, and in a time-series setting the sliding window of size 20 with stride 3 can create overlapping windows and temporal leakage. The central claim of the paper therefore rests on an invalid evaluation.
[Section V, Conclusion] The conclusion states that the CNN-GRU prototype demonstrates 'an almost 20% enhancement in precision over conventional means.' No table, figure, or equation in the manuscript supports this number. Table I compares CNN+GRU to standalone CNN and GRU on the training set only, and the text does not define what 'conventional means' includes. This claim must be removed or substantiated with held-out evaluation and explicit baselines.
[Sections III.C and IV.D] The Introduction and Conclusion repeatedly credit an 'attention-guided mechanism' as part of the model, but the architecture described in Section III.C and the GRU equations (3)-(6) contain no attention layer, and the experiments in Section IV.D make no mention of attention. Either the model must be described with its attention component, or the claims about attention must be deleted.
[Section IV.D] The paper compares CNN+GRU only to standalone CNN and standalone GRU in Table I. The text in Sections II and V mentions LSTM, ARIMA, 'traditional statistical methods,' and other baselines, but none of these are ever evaluated or reported. Without these comparisons, the claim that the hybrid maintains 'apex predictive potency' relative to LSTM or recurrent architectures is unsupported.
[Section IV.B and IV.D] The paper asserts that sentiment categories derived from social media and financial news text predict future S&P 500 values, but this link is never tested. Section IV.B only describes synchronizing sentiment labels with market metrics, and Section IV.D directly trains the model without any analysis of the lag structure or predictive relationship between sentiment and future returns. The entire prediction claim is therefore built on an unvalidated assumption.

minor comments (5)

[Section II, References] Several references do not support the claims attached to them: reference [1] is described as a CNN-LSTM hybrid but the cited title concerns extreme learning machines, and reference [17] is described as a CNN-GRU study but the cited title is 'Temporal convolutional networks for stock trend prediction.' These citations should be corrected or replaced.
[Equations (1)-(6)] The notation for the learning rate, loss function, and weight matrices is inconsistent across the paper, and equations (3)-(6) use Unicode-style formatting that will not render correctly in a standard LaTeX submission. The loss function in Section IV.D is described as a combination of MSE and cross-entropy, but this is not stated in the theoretical sections.
[Section IV.A] The dataset description gives no exact date range, number of samples, number of stocks, or class distribution. The phrase 'a selection spanning half a decade' is too vague to assess the experimental design or reproducibility.
[Section IV.B] The 'linked data model' used for sentiment categorization is never identified by name, version, or training corpus. Readers cannot know what sentiment labels mean or how reliable they are.
[General] The manuscript contains numerous language and formatting issues, including 'Tabular Table I,' 'prognostic yield,' 'bourse prognostics,' and inconsistent capitalization of 'CNN+GRU' versus 'CNN-GRU.' A thorough editorial pass is needed.

Circularity Check

1 steps flagged · score 6.0 of 10

Central '~20% precision enhancement' is computed on the training set, so the claimed predictive improvement reduces to in-sample fit rather than out-of-sample prediction.

fitted input called prediction [Section IV.D, Table I; Section V Conclusion]
"TABLE I. COMPARISON OF PREDICTION PERFORMANCE OF DIFFERENT MODELS ON THE TRAINING SET ... More concretely, the CNN-GRU prototype excels ... demonstrating an almost 20% enhancement in precision over conventional means, a feat credited to its profound comprehension and assimilation of intricate emotional progression patterns."

Section IV.C states that 'The dataset is divided into three subsets: training, validation, and test sets,' but Section IV.D reports no validation or test metrics. The only quantitative support for the central claim is Table I, whose caption explicitly labels it 'ON THE TRAINING SET.' The CNN-GRU and baseline models are fitted on that same training set, so the precision, recall, and F1 values in Table I are measures of in-sample fit, not held-out predictions. The conclusion then converts those training-set numbers into the paper's headline predictive result ('almost 20% enhancement in precision over conventional means').

full rationale

The paper contains no mathematical derivation whose conclusion is equivalent to its premises: the GRU equations in Section III are standard, and the CNN+GRU architecture is a known combination described rather than derived. The self-references that are likely attributable to the authors (e.g., Lin et al. [10]) are contextual or motivational, not load-bearing: no uniqueness theorem or ansatz is imported from them, and the core experimental claim does not depend on a cited prior result. The one genuine circularity is in the evaluation logic: Section IV.C announces a train/validation/test split, Section IV.D reports only training-set metrics in Table I, and the Conclusion presents those metrics as a demonstrated predictive improvement of 'almost 20%' in precision. That is the pattern 'fitted input called prediction'—a model fitted to the training data is then credited with predictive power based on the same training data, with no held-out validation or test numbers. This is partial circularity of the central claim rather than a defect in the architecture description or in the standard equations, so the score is 6 rather than higher; the paper is also weak on external validity, but that is a correctness concern beyond the circularity analysis.

Assumptions & free parameters 6 free parameters · 3 assumptions · 0 invented entities

The model relies on standard deep learning components and a pre-trained sentiment labeler; no new physical or conceptual entity is introduced. The load-bearing assumptions are the predictive link between text sentiment and market prices, the reliability of the sentiment pre-classifier, and the representativeness of the data split.

free parameters (6)

learning_rate = 0.0001
Chosen by hand; no tuning procedure described in Section IV.C.
batch_size = 50
Chosen by hand; no tuning procedure described in Section IV.C.
sliding_window_size = 20
Chosen by hand; no tuning procedure described in Section IV.C.
convolutional_stride = 3
Chosen by hand; no tuning procedure described in Section IV.C.
gru_hidden_nodes = 32
Chosen by hand; no tuning procedure described in Section IV.C.
joint_loss_weighting
The combined MSE and cross-entropy loss requires a weighting term, but the paper never specifies it in Section IV.D.

assumptions (3)

domain assumption Sentiment from social media and news text predicts S&P 500 price movements.
Assumed in Section IV.A-B; sentiment labels are synchronized with market metrics but no statistical evidence of predictiveness is given.
domain assumption The pre-trained linked data model [24] correctly classifies text sentiment.
Invoked in Section IV.B as the sentiment labeling step; no validation of this model on the target text is reported.
domain assumption The training set is representative of future market conditions.
The train/validation/test split is mentioned in Section IV.C, but no date ranges or market regime details are provided, so generalization is assumed.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Integrative Analysis of Financial Market Sentiment Using CNN and GRU for Risk Prediction and Alert Systems." pith.science (2026). https://pith.science/paper/4HVRDGEN

@misc{pith2026241210199,
  author       = {Pith},
  title        = {Pith review of: Integrative Analysis of Financial Market Sentiment Using CNN and GRU for Risk Prediction and Alert Systems},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4HVRDGEN}},
  note         = {Machine review of arXiv:2412.10199}
}

read the original abstract

This document presents an in-depth examination of stock market sentiment through the integration of Convolutional Neural Networks (CNN) and Gated Recurrent Units (GRU), enabling precise risk alerts. The robust feature extraction capability of CNN is utilized to preprocess and analyze extensive network text data, identifying local features and patterns. The extracted feature sequences are then input into the GRU model to understand the progression of emotional states over time and their potential impact on future market sentiment and risk. This approach addresses the order dependence and long-term dependencies inherent in time series data, resulting in a detailed analysis of stock market sentiment and effective early warnings of future risks.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

Optimizing Large Language Models with an Enhanced LoRA Fine-Tuning Algorithm for Efficiency and Robustness in NLP Tasks
cs.CL 2024-12 reject novelty 2.0 of 10

A modified LoRA update with per-matrix learning rates and an object-detection-style density term is reported to slightly improve QQP accuracy over GPT-4 baselines.
Leveraging Convolutional Neural Network-Transformer Synergy for Predictive Modeling in Risk-Based Applications
q-fin.RM 2024-12 reject novelty 2.0 of 10

A CNN+Transformer hybrid is reported to outperform tree-based baselines on credit default prediction, with accuracy 0.8197 and AUC 0.7921.

Reference graph

Works this paper leans on

25 extracted references · 23 canonical work pages · cited by 2 Pith papers

[17]

Temporal convolutional networks for stock trend prediction,

F. Zhou, H. Zhou, Z. Yang, and L. Yang, "Temporal convolutional networks for stock trend prediction," Expert Systems with Applications, vol. 189, p. 116078, 2022

work page 2022
[23]

Forecasting stock market indices using the recurrent neural network based hybrid models: CNN- LSTM, GRU - CNN, and ensemble models,

H. Song and H. Choi, "Forecasting stock market indices using the recurrent neural network based hybrid models: CNN- LSTM, GRU - CNN, and ensemble models," Applied Sciences, vol. 13, no. 7, p. 4644, 2023

work page 2023
[1]

Empirical analysis: Stock market prediction via extreme learning machine,

X. Li, H. Xie, R. Wang, Y. Cai, J. Cao, F. Wang, H. Min, and X. Deng, "Empirical analysis: Stock market prediction via extreme learning machine," Neural Computing and Applications, vol. 28, no. 1, pp. 67- 78, 2017

work page 2017
[2]

A novel CNN-GRU-based hybrid approach for short-term residential load forecasting,

M. Sajjad, Z. A. Khan, A. Ullah, T. Hussain, W. Ullah, M. Y. Lee, and S. W. Baik, "A novel CNN-GRU-based hybrid approach for short-term residential load forecasting," IEEE Access, vol. 8, pp. 143759-143768, 2020

work page 2020
[3]

Big data and machine learning in defence,

Y. Weng and J. Wu, "Big data and machine learning in defence," International Journal of Computer Science and Information Technology, vol. 16, no. 2, pp. 25-35, 2024

work page 2024
[4]

Mopir: a machine-learning- oriented data preprocessing pipeline for precise analysis of infrared spectroscopy,

Z. Huang et al., "Mopir: a machine-learning- oriented data preprocessing pipeline for precise analysis of infrared spectroscopy,"

work page
[5]

Fortifying the global data fortress: a multidimensional examination of cyber security indexes and data protection measures across 193 nations,

Y. Weng and J. Wu, "Fortifying the global data fortress: a multidimensional examination of cyber security indexes and data protection measures across 193 nations," International Journal of Frontiers in Engineering Technology, vol. 6, no. 2, pp. 13-28, 2024

work page 2024
[6]

Advanced Financial Fraud Detection Using GNN-CL Model

Y. Cheng et al., "Advanced Financial Fraud Detection Using GNN-CL Model," arXiv preprint arXiv:2407.06529, 2024

work page Pith review arXiv 2024

Show all 25 references

[7]

Application of Black -Litterman Bayesian in Statistical Arbitrage,

Q. Zhou, "Application of Black -Litterman Bayesian in Statistical Arbitrage," arXiv preprint arXiv:2406.06706, 2024

2024 arXiv
[8]

Portfolio Optimization with Robust Covariance and Conditional Value-at- Risk Constraints,

Q. Zhou, "Portfolio Optimization with Robust Covariance and Conditional Value-at- Risk Constraints," arXiv preprint arXiv:2406.00610, 2024

2024 arXiv
[9]

GraphCNNpred: A stock market indices prediction using a Graph based deep learning system,

Y. Jin, "GraphCNNpred: A stock market indices prediction using a Graph based deep learning system," arXiv preprint arXiv:2407.03760, 2024

2024 arXiv
[10]

Text Sentiment Detection and Classification Based on Integrated Learning Algorithm,

Z. Lin et al., "Text Sentiment Detection and Classification Based on Integrated Learning Algorithm," Applied Science and Engineering Journal for Advanced Research, vol. 3, no. 3, pp. 27-33, 2024

2024
[11]

Optimization Strategies for Self -Supervised Learning in the Use of Unlabeled Data,

H. Zhao et al., "Optimization Strategies for Self -Supervised Learning in the Use of Unlabeled Data," Journal of Theory and Practice of Engineering Science, vol. 4, no. 05, pp. 30-39, 2024

2024
[12]

Economic prediction using neural networks: The case of IBM daily stock returns,

H. White, "Economic prediction using neural networks: The case of IBM daily stock returns," in IEEE International Conference on Neural Networks, 1988, vol. 2, pp. 451-458

1988
[13]

Forecasting stock markets using wavelet transforms and recurrent neural networks: An integrated system based on artificial bee colony algorithm,

T. J. Hsieh, H. F. Hsiao, and W. C. Yeh, "Forecasting stock markets using wavelet transforms and recurrent neural networks: An integrated system based on artificial bee colony algorithm," Applied Soft Computing, vol. 11, no. 2, pp. 2510-2525, 2011

2011
[14]

Deep learning for multivariate financial time series,

G. Batres -Estrada, "Deep learning for multivariate financial time series," KTH Royal Institute of Technology, 2015

2015
[15]

Stock market's price movement prediction with LSTM neural networks,

D. M. Nelson, A. C. Pereira, and R. A. de Oliveira, "Stock market's price movement prediction with LSTM neural networks," in International Joint Conference on Neural Networks (IJCNN), 2017, pp. 1419-1426

2017
[16]

A comparison of ARIMA and LSTM in forecasting time series,

S. Siami-Namini, N. Tavakoli, and A. S. Namin, "A comparison of ARIMA and LSTM in forecasting time series," in 17th IEEE International Conference on Machine Learning and Applications (ICMLA), 2018, pp. 1394-1401

2018
[18]

Attention is all you need,

A. Vaswani et al., "Attention is all you need," in Advances in Neural Information Processing Systems, vol. 30, 2017

2017
[19]

Transformers for stock market prediction,

F. Pasch, "Transformers for stock market prediction," arXiv preprint arXiv:2205.13504, 2022

2022 arXiv
[20]

Forecasting of currency exchange rates using an adaptive ARMA model with differential evolution based training,

M. Rout, B. Majhi, R. Majhi, and G. Panda, "Forecasting of currency exchange rates using an adaptive ARMA model with differential evolution based training," Journal of King Saud University-Computer and Information Sciences, vol. 26, no. 1, pp. 7-18, 2013

2013
[21]

Survival prediction across diverse cancer types using neural networks,

X. Yan, W. Wang, M. Xiao, Y. Li, and M. Gao, "Survival prediction across diverse cancer types using neural networks," in Proceedings of the 2024 7th International Conference on Machine Vision and Applications, 2024, pp. 134-138

2024
[22]

Gate-variants of gated recurrent unit (GRU) neural networks,

R. Dey and F. M. Salem, "Gate-variants of gated recurrent unit (GRU) neural networks," in 2017 IEEE 60th International Midwest Symposium on Circuits and Systems (MWSCAS), 2017, pp. 1597- 1600

2017
[24]

Investigation of creating accessibility linked data based on publicly available accessibility datasets,

Y. Li, X. Yan, M. Xiao, W. Wang, and F. Zhang, "Investigation of creating accessibility linked data based on publicly available accessibility datasets," in Proceedings of the 2023 13th International Conference on Communication and Network Security, 2023, pp. 77- 81

2023
[2024]

Available: http s://doi.org/10.21203/rs.3.rs- 4477355/v1

[Online]. Available: http s://doi.org/10.21203/rs.3.rs- 4477355/v1

doi

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.

[1] [17]

Temporal convolutional networks for stock trend prediction,

F. Zhou, H. Zhou, Z. Yang, and L. Yang, "Temporal convolutional networks for stock trend prediction," Expert Systems with Applications, vol. 189, p. 116078, 2022

work page 2022

[2] [23]

Forecasting stock market indices using the recurrent neural network based hybrid models: CNN- LSTM, GRU - CNN, and ensemble models,

H. Song and H. Choi, "Forecasting stock market indices using the recurrent neural network based hybrid models: CNN- LSTM, GRU - CNN, and ensemble models," Applied Sciences, vol. 13, no. 7, p. 4644, 2023

work page 2023

[3] [1]

Empirical analysis: Stock market prediction via extreme learning machine,

X. Li, H. Xie, R. Wang, Y. Cai, J. Cao, F. Wang, H. Min, and X. Deng, "Empirical analysis: Stock market prediction via extreme learning machine," Neural Computing and Applications, vol. 28, no. 1, pp. 67- 78, 2017

work page 2017

[4] [2]

A novel CNN-GRU-based hybrid approach for short-term residential load forecasting,

M. Sajjad, Z. A. Khan, A. Ullah, T. Hussain, W. Ullah, M. Y. Lee, and S. W. Baik, "A novel CNN-GRU-based hybrid approach for short-term residential load forecasting," IEEE Access, vol. 8, pp. 143759-143768, 2020

work page 2020

[5] [3]

Big data and machine learning in defence,

Y. Weng and J. Wu, "Big data and machine learning in defence," International Journal of Computer Science and Information Technology, vol. 16, no. 2, pp. 25-35, 2024

work page 2024

[6] [4]

Mopir: a machine-learning- oriented data preprocessing pipeline for precise analysis of infrared spectroscopy,

Z. Huang et al., "Mopir: a machine-learning- oriented data preprocessing pipeline for precise analysis of infrared spectroscopy,"

work page

[7] [5]

Fortifying the global data fortress: a multidimensional examination of cyber security indexes and data protection measures across 193 nations,

Y. Weng and J. Wu, "Fortifying the global data fortress: a multidimensional examination of cyber security indexes and data protection measures across 193 nations," International Journal of Frontiers in Engineering Technology, vol. 6, no. 2, pp. 13-28, 2024

work page 2024

[8] [6]

Advanced Financial Fraud Detection Using GNN-CL Model

Y. Cheng et al., "Advanced Financial Fraud Detection Using GNN-CL Model," arXiv preprint arXiv:2407.06529, 2024

work page Pith review arXiv 2024