Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Multi-Level Sentiment Analysis Framework for Financial Texts

Yiwei Liu, Junbo Wang, Lei Long, Xin Li, Ruiting Ma, Yuankai Wu, Xuebin Chen

TL;DR

The paper tackles the limitation of single-level sentiment analysis in finance by proposing a multi-level framework that combines micro-level firm-specific ABSA and meso-level industry-specific SLSA, augmented with a duration-aware smoothing mechanism. Using a Chinese bond market corpus (2013–2023, ~1.39M texts), it builds a composite sentiment index that, when added to a rolling-window bond default risk forecasting model, reduces forecasting errors significantly (MAE by ~3.25% and MAPE by ~10.97%). The approach leverages a knowledge graph and RAG to propagate industry signals, and applies wavelet smoothing to capture latency and persistence in sentiment diffusion, with empirical evidence linking sentiment to social risk events. The framework demonstrates that integrated, temporally-aware sentiment provides a more nuanced understanding of risk dynamics and improves practical forecasting performance in bond markets, with potential extensions to other asset classes.

Abstract

Existing financial sentiment analysis methods often fail to capture the multi-faceted nature of risk in bond markets due to their single-level approach and neglect of temporal dynamics. We propose Multi-Level Sentiment Analysis based on pre-trained language models (PLMs) and large language models (LLMs), a novel framework that systematically integrates firm-specific micro-level sentiment, industry-specific meso-level sentiment, and duration-aware smoothing to model the latency and persistence of textual impact. Applying our framework to the comprehensive Chinese bond market corpus constructed by us (2013-2023, 1.39M texts), we extracted a daily composite sentiment index. Empirical results show statistically measurable improvements in credit spread forecasting when incorporating sentiment (3.25% MAE and 10.96% MAPE reduction), with sentiment shifts closely correlating with major social risk events and firm-specific crises. This framework provides a more nuanced understanding of sentiment across different market levels while accounting for the temporal evolution of sentiment effects.

A Multi-Level Sentiment Analysis Framework for Financial Texts

TL;DR

The paper tackles the limitation of single-level sentiment analysis in finance by proposing a multi-level framework that combines micro-level firm-specific ABSA and meso-level industry-specific SLSA, augmented with a duration-aware smoothing mechanism. Using a Chinese bond market corpus (2013–2023, ~1.39M texts), it builds a composite sentiment index that, when added to a rolling-window bond default risk forecasting model, reduces forecasting errors significantly (MAE by ~3.25% and MAPE by ~10.97%). The approach leverages a knowledge graph and RAG to propagate industry signals, and applies wavelet smoothing to capture latency and persistence in sentiment diffusion, with empirical evidence linking sentiment to social risk events. The framework demonstrates that integrated, temporally-aware sentiment provides a more nuanced understanding of risk dynamics and improves practical forecasting performance in bond markets, with potential extensions to other asset classes.

Abstract

Existing financial sentiment analysis methods often fail to capture the multi-faceted nature of risk in bond markets due to their single-level approach and neglect of temporal dynamics. We propose Multi-Level Sentiment Analysis based on pre-trained language models (PLMs) and large language models (LLMs), a novel framework that systematically integrates firm-specific micro-level sentiment, industry-specific meso-level sentiment, and duration-aware smoothing to model the latency and persistence of textual impact. Applying our framework to the comprehensive Chinese bond market corpus constructed by us (2013-2023, 1.39M texts), we extracted a daily composite sentiment index. Empirical results show statistically measurable improvements in credit spread forecasting when incorporating sentiment (3.25% MAE and 10.96% MAPE reduction), with sentiment shifts closely correlating with major social risk events and firm-specific crises. This framework provides a more nuanced understanding of sentiment across different market levels while accounting for the temporal evolution of sentiment effects.

Paper Structure

This paper contains 27 sections, 15 equations, 10 figures, 8 tables, 3 algorithms.

Table of Contents

  1. Introduction
  2. Related Works
  3. Methodology
  4. Terminology And Task Decomposition
  5. Datasets in Task 1
  6. Labeled Sentiment Corpus $\mathcal{D}_{1}$
  7. Knowledge Graph $\mathcal{G}$ And Knowledge Base $\mathcal{B}$
  8. Large-scale Unlabeled Corpus $\mathcal{D}_{2}$
  9. Firms Features Dataset $\mathcal{D}_{3}$
  10. Firm-Specific Sentiment Analysis in Task 2
  11. Industry-Specific Sentiment Analysis in Task 2
  12. Aggregation and Duration Function in Task 2
  13. Bond Default Risk Forecasting Modeling in Task 3
  14. Expriments
  15. Setting
  16. ...and 12 more sections

Figures (10)

  • Figure 1: Task Decomposition. Each task corresponds to one of our contributions in Section. \ref{['sec:intro']}.
  • Figure 2: ABSA Procedure.
  • Figure 3: SLSA Procedure.
  • Figure 4: The Transformation Relationships.
  • Figure 5: Visualization of Rolling Window.
  • ...and 5 more figures