Dissecting Corporate Reputation Risk in News Media

Overview

Corporate reputations are built and destroyed in the media. This project (March–September 2023) developed an aspect-based sentiment analysis system that goes beyond simple positive/negative classification to understand which specific aspects of a company or CEO’s reputation are being affected by news coverage.

How It Works

Traditional sentiment analysis treats a news article as having a single overall sentiment. But a single article might praise a CEO’s innovation strategy while criticizing the company’s environmental record. Our approach breaks down each article into distinct reputation aspects — such as leadership, product quality, social responsibility, financial performance, and governance — and assigns sentiment scores to each.

This granular analysis enables a much more nuanced view of reputation dynamics than aggregate approaches.

Technical Approach

We combined Korean NLP preprocessing with fine-tuned transformer models for aspect extraction and sentiment classification. The system processes Korean news articles, identifies mentioned entities (companies and executives), extracts reputation-relevant aspects, and classifies sentiment at the aspect level.

Applications

The resulting tool can serve as an early warning system for reputation risks, helping communications teams identify emerging issues before they escalate. It can also be used for competitive intelligence, tracking how a company’s reputation compares to peers across specific dimensions.

Collaborators

• Cheju Halla University: Lead research institution
• Industry Partners: Data access and validation