Fusing Financial News Aspect-Attention via DeBERTa-v3 with Edge-Cloud Hybrid Deep Learning for Real-Time Market Liquidity and Asset Price Forecasting

• Billy Elly

  Lautech

  Author

The accurate forecasting of market liquidity and asset prices remains a formidable challenge in quantitative finance, primarily due to the inherent volatility, non-linearity, and multi-faceted nature of financial markets. Traditional models often fail to capture the complex interplay between unstructured textual news data and structured market indicators, while also struggling with the latency requirements of real-time applications. This research addresses this gap by proposing a novel hybrid framework that fuses aspect-based sentiment analysis from financial news with a deep learning architecture for enhanced predictive performance. The methodology integrates a DeBERTa-v3 transformer model, enhanced with aspect-attention mechanisms and contrastive learning, to extract nuanced, entity-specific sentiment signals from financial news streams. These extracted sentiment features are subsequently fused with technical indicators and market microstructure data within a hybrid CNN-LSTM network, deployed on an edge-cloud architecture to balance computational efficiency with low-latency inference. The empirical evaluation demonstrates that the proposed framework achieves superior predictive accuracy, attaining a Mean Absolute Error (MAE) of 0.018 and an R² score of 0.974 for asset price forecasting, outperforming benchmark models such as standalone LSTM and ARIMA. The model also demonstrates robust performance in predicting liquidity shifts, with an F1-score of 89.4% for directional movement classification, offering a sgnificant advancement for real-time financial decision support systems.

• pdf

Copyright (c) 2026 Billy Elly (Author)

This work is licensed under a Creative Commons Attribution 4.0 International License.