Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Personalized Chain-of-Thought Summarization of Financial News for Investor Decision Support

Tianyi Zhang, Mu Chen

TL;DR

The paper tackles information overload in financial news by proposing a personalized chain-of-thought (CoT) summarization framework that uses keyword-driven filtering to produce event-driven outputs tailored to investor needs. It introduces a four-stage pipeline—robust text extraction, initial financial summarization, metadata-enhanced refinement via few-shot learning, and personalized response generation with three investment actions. Automated metrics show large gains (BLEU 0.1786, ROUGE-L 0.4028; +267% and +90% over GPT-4o), while professional analysts prefer the enhanced outputs, and a binary relevance strategy improves personalization accuracy by about 40% over ranking-based methods. The work demonstrates that domain-grounded CoT with professional guidance can deliver accurate, actionable investor insights and can be extended to real-time data and other decision-support domains.

Abstract

Financial advisors and investors struggle with information overload from financial news, where irrelevant content and noise obscure key market signals and hinder timely investment decisions. To address this, we propose a novel Chain-of-Thought (CoT) summarization framework that condenses financial news into concise, event-driven summaries. The framework integrates user-specified keywords to generate personalized outputs, ensuring that only the most relevant contexts are highlighted. These personalized summaries provide an intermediate layer that supports language models in producing investor-focused narratives, bridging the gap between raw news and actionable insights.

Personalized Chain-of-Thought Summarization of Financial News for Investor Decision Support

TL;DR

The paper tackles information overload in financial news by proposing a personalized chain-of-thought (CoT) summarization framework that uses keyword-driven filtering to produce event-driven outputs tailored to investor needs. It introduces a four-stage pipeline—robust text extraction, initial financial summarization, metadata-enhanced refinement via few-shot learning, and personalized response generation with three investment actions. Automated metrics show large gains (BLEU 0.1786, ROUGE-L 0.4028; +267% and +90% over GPT-4o), while professional analysts prefer the enhanced outputs, and a binary relevance strategy improves personalization accuracy by about 40% over ranking-based methods. The work demonstrates that domain-grounded CoT with professional guidance can deliver accurate, actionable investor insights and can be extended to real-time data and other decision-support domains.

Abstract

Financial advisors and investors struggle with information overload from financial news, where irrelevant content and noise obscure key market signals and hinder timely investment decisions. To address this, we propose a novel Chain-of-Thought (CoT) summarization framework that condenses financial news into concise, event-driven summaries. The framework integrates user-specified keywords to generate personalized outputs, ensuring that only the most relevant contexts are highlighted. These personalized summaries provide an intermediate layer that supports language models in producing investor-focused narratives, bridging the gap between raw news and actionable insights.

Paper Structure

This paper contains 33 sections, 6 equations, 5 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Financial News Summarization and Domain Adaptation
  4. Personalized and Controllable Summarization
  5. Reasoning and Chain-of-Thought in Summarization
  6. Methodology
  7. Data Processing and Text Extraction
  8. Text Extraction Pipeline
  9. English Language Filtering
  10. Metadata Preservation
  11. Initial Financial Summarization
  12. Stage 3: Enhanced Summary Processing with Few-Shot Learning
  13. LLM-Based Metadata Extraction
  14. Few-Shot Learning for Financial Summary Enhancement
  15. Stage 4: Personalized Response Generation
  16. ...and 18 more sections

Figures (5)

  • Figure 1: Overview of Personalized Chain-of-Thought Financial News Summarization Framework
  • Figure 2: Second Stage: Initial Summarization Prompt Template and Sample Output
  • Figure 3: Third Stage: Enhanced Summary Few-Shot Learning Prompt Template
  • Figure 4: Fourth Stage: Personalized Response Generation Sample Output
  • Figure 5: Sample Summary Comparison