A Multi-Level Sentiment Analysis Framework for Financial Texts

2025-04-03

AI Summary

Existing bond sentiment analysis predominantly relies on single-layer static modeling, failing to capture the multidimensionality and temporal lag of risk. To address this, we propose a micro–meso–temporal tri-level sentiment coupling framework: (i) micro-level integration of corporate textual data; (ii) meso-level embedding of industry dynamics; and (iii) temporal-level incorporation of duration-aware mechanisms to model the delay and persistence of sentiment effects. Leveraging a self-constructed corpus of 1.39 million Chinese bond-related documents (2013–2023), we combine pre-trained language models (PLMs) and large language models (LLMs) to construct a daily composite sentiment index. Experiments demonstrate that this index reduces mean absolute error (MAE) in credit spread prediction by 3.25% and mean absolute percentage error (MAPE) by 10.96%. Moreover, sentiment spikes exhibit significant responsiveness to societal risk events and corporate crises. This work achieves, for the first time, multi-scale, high-temporal-resolution, and interpretable sentiment modeling for the bond market.

Technology Category

• Natural Language Processing: Sentiment Analysis, Stylistic Analysis, and Argument Mining
• Data Mining & Knowledge Management: Graph Mining, Social Network Analysis & Community
• Cognitive Modeling & Cognitive Systems: Social Cognition And Interaction

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Web Mining and Content Analysis: Sentiment analysis and opinion mining
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

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.

Problem

• Existing methods miss multi-faceted bond market risk sentiment.
• Proposed framework integrates micro, meso, and temporal sentiment levels.
• Improves credit spread forecasting with measurable accuracy gains.

Innovation

• Multi-Level Sentiment Analysis with PLMs and LLMs
• Micro, meso, and duration-aware sentiment integration
• Daily composite sentiment index for credit forecasting