From News to Trends: A Financial Time Series Forecasting Framework with LLM-Driven News Sentiment Analysis and Selective State Spaces DOI
Renjie Wang, Minghui Sun, Limin Wang

et al.

Research Square (Research Square), Journal Year: 2025, Volume and Issue: unknown

Published: April 17, 2025

Abstract Stock price prediction is inherently challenging due to market volatility and the influence of external factors. Traditional forecasting methods primarily rely on historical data, limiting their ability capture sentiment embedded in financial news. To address this limitation, we propose Senti-MambaMoE, a novel model that integrates stock prices with information extracted from Specifically, fine-tune DeepSeek-based large language (LLM) for classification incorporate into our predictive framework. At core approach MambaMoE, which leverages efficiency state space models (SSMs) long-range dependencies while maintaining linear computational complexity, making it well-suited time series forecasting. Additionally, MoE mechanism improves model’s diverse behaviors by dynamically selecting specialized experts based data patterns. Experimental results demonstrate Senti-MambaMoE outperforms LSTM-based 23.7% Transformer-based 6.3%, highlighting its superior performance short-term prediction.

Language: Английский

Pattern-oriented Attention Mechanism for Multivariate Time Series Forecasting DOI Open Access
Hanwen Hu, Zhangchi Han, Shiyou Qian

et al.

ACM Transactions on Knowledge Discovery from Data, Journal Year: 2025, Volume and Issue: unknown

Published: Jan. 17, 2025

Multivariate time series forecasting is applied in many domains, such as finance, transportation and industry. The main challenge of precise lies accurately capturing latent dependencies. Recent studies develop various frameworks to reduce computational complexity or enhance the learning intricate relationships, while lacking interpretability generality. In this paper, we aim elucidate capture dependencies recognition patterns. We believe that patterns can be formally described from two aspects: shapes segments frequently repeat, corresponding forms repetitions. Drawing upon idea, design a multivariate model named PRformer 1 , which incorporates pattern-oriented attention mechanism pattern-based projector. perceive different repetitions by embedded with similarity evaluation metrics between segments, filter out noise extract potential statistical-driven weighting scheme. projector employed form results deriving representative set ones. By incorporating explicit definitions patterns, interpretable general scenarios. Experimental on seven datasets demonstrate outperforms six state-of-the-art models about 10.7% accuracy.

Language: Английский

Citations

0
Real-time prediction of port water levels based on EMD-PSO-RBFNN DOI Creative Commons
Lijun Wang, Shenghao Liao, Sisi Wang

et al.

Frontiers in Marine Science, Journal Year: 2025, Volume and Issue: 12

Published: Jan. 23, 2025

Addressing the spatial variability, temporal dynamics, and non-linearity characteristics of port water levels, a hybrid prediction scheme was proposed, which integrates empirical mode decomposition (EMD) with radial basis function neural network (RBFNN), optimized using particle swarm optimization (PSO) algorithm. First, through application EMD, level time series decomposed into sub-series characterized by lower non-linearity. Subsequently, PSO applied to fine-tune center spread parameters RBFNN, thereby enhancing model’s predictive performance. The PSO-RBFNN model employed make predictions on sub-series. Finally, reconstruction predicted yielded final predictions. feasibility effectiveness proposed were validated measured data. Results from simulations highlighted ability deliver accurate across various lead times. Furthermore, comparative analysis revealed that outperforms alternative methods in prediction. Therefore, serves as reliable, efficient, real-time tool, providing robust support for operational safety.

Language: Английский

Citations

0
Enhancing Music Audio Signal Recognition through CNN-BiLSTM Fusion with De-noising Autoencoder for Improved Performance DOI
Xin Mao, Ye Tian, Tao Jin

et al.

Neurocomputing, Journal Year: 2025, Volume and Issue: unknown, P. 129607 - 129607

Published: Feb. 1, 2025

Language: Английский

Citations

0
PyPortOptimization: A Portfolio Optimization Pipeline Leveraging Multiple Expected Return Methods, Risk Models, and Post-Optimization Allocation Techniques DOI Creative Commons

Rushikesh Nakhate,

Harikrishnan Ramachandran,

Anjali Mahajan

et al.

MethodsX, Journal Year: 2025, Volume and Issue: 14, P. 103211 - 103211

Published: Feb. 7, 2025

Language: Английский

Citations

0
Beyond spatial neighbors: Utilizing multivariate transfer entropy for interpretable graph-based spatio–temporal forecasting DOI
Safaa Berkani, Adil Bahaj, Bassma Guermah

et al.

Engineering Applications of Artificial Intelligence, Journal Year: 2025, Volume and Issue: 146, P. 110161 - 110161

Published: Feb. 17, 2025

Language: Английский

Citations

0
Feature-driven hybrid attention learning for accurate water quality prediction DOI
Xuan Yao, Zeshui Xu, Tianyu Ren

et al.

Expert Systems with Applications, Journal Year: 2025, Volume and Issue: unknown, P. 127160 - 127160

Published: March 1, 2025

Language: Английский

Citations

0
Forecasting stock market time series through the integration of bee colony optimizer and multivariate empirical mode decomposition with extreme gradient boosting regression DOI
Xuefeng Liu, Zhixin Wu,

Jiayue Xin

et al.

Engineering Applications of Artificial Intelligence, Journal Year: 2025, Volume and Issue: 148, P. 110353 - 110353

Published: March 10, 2025

Language: Английский

Citations

0
Stock Price Forecasting With Integration of Sectoral Behavior: A Deep Auto‐Optimized Multimodal Framework DOI Open Access

Renu Saraswat,

Ajit Kumar

Journal of Forecasting, Journal Year: 2025, Volume and Issue: unknown

Published: March 16, 2025

ABSTRACT This study proposes a novel deep auto‐optimized architecture for stock price forecasting that integrates sectoral behavior with individual sentiment to improve predictive accuracy. Traditional prediction models often focus solely on behavior, overlooking the impact of broader trends. The proposed approach utilizes advanced learning models, including gated recurrent units (GRU), bidirectional GRU, long short‐term memory (LSTM), and LSTM, their hybrid ensembles. These are built using Keras functional API auto ML network search technology. current multimodal framework incorporates significantly improving performance metrics. research highlights critical role integrating in models.

Language: Английский

Citations

0
Time–Frequency Domain Joint Noise Reduction Multi-Resolution Power System Time Series Prediction Network DOI
Zhaofeng Cao,

Hangwei Tian,

Qihao Xu

et al.

Computers & Electrical Engineering, Journal Year: 2025, Volume and Issue: 123, P. 110255 - 110255

Published: March 20, 2025

Language: Английский

Citations

0
A new perspective on non-ferrous metal price forecasting: An interpretable two-stage ensemble learning-based interval-valued forecasting system DOI

Wendong Yang,

Hao Zhang, Jianzhou Wang

et al.

Advanced Engineering Informatics, Journal Year: 2025, Volume and Issue: 65, P. 103267 - 103267

Published: March 21, 2025

Language: Английский

Citations

0