RA-FinBERT: Rule-aware LoRA adaptation for low-resource financial sentiment classification
2026-08-10 • Computation and Language
Computation and LanguageMachine Learning
AI summaryⓘ
The authors created a new model called RA-FinBERT to better understand the sentiment in financial news. They combined a popular language model (FinBERT) with simple, extra information from sentiment rules and source details without adding many extra calculations. This made the model more accurate at guessing if news is positive, negative, or neutral, especially improving how often it correctly recognized neutral news. Their approach works well even on regular computers, making it useful when resources are limited. Overall, they showed that mixing smart rules with advanced language models helps improve financial sentiment analysis.
Financial sentiment analysisFinBERTLow-rank adaptation (LoRA)VADER sentiment analysisMetadata featuresMacro F1 scoreNeutral-class recallParameter-efficient modelsDistilBERTText classification
Authors
Fan Zhang, Jiaming Li
Abstract
Financial sentiment analysis converts unstructured financial news into quantitative signals that can support market analysis and decision-making. Existing work on resource-efficient financial NLP has largely focused on compressing or adapting pretrained language models, with less attention to combining contextual representations with lightweight rule-derived features. This study develops Rule-Aware FinBERT (RA-FinBERT), a parameter-efficient framework that integrates low-rank adaptation (LoRA) with three continuous VADER-derived sentiment proportions (positive, negative, and neutral) and a source-level metadata feature. The standardized four-dimensional feature vector is directly concatenated with the 768-dimensional final-layer FinBERT [CLS] representation and passed through a lightweight classification head. This design introduces only 1,024 additional trainable weights relative to a structurally matched text-only FinBERT model. RA-FinBERT was evaluated against text-only FinBERT and a lightweight DistilBERT baseline for three-class sentiment classification of financial-news titles and descriptions. On the held-out test set, RA-FinBERT achieved 69.89% accuracy and a macro F1 score of 0.634, compared with 63.44% and 0.526 for text-only FinBERT. Neutral-class recall increased from 18.18% to 45.45%. The framework supports both CPU and GPU execution, offering a lightweight and practical approach to financial sentiment classification under constrained computational resources. These findings indicate that rule-derived sentiment information and source metadata can provide complementary signals to contextual FinBERT representations and improve performance with minimal additional model complexity.