Measuring Sentiment News with Transformer-Based Language Models

2026-07-15

AI Summary

Traditional dictionary-based approaches struggle to accurately capture contextual nuances, negation, and semantic structures in financial news, often yielding sentiment assessments that diverge from human judgment. This study proposes a sentence-level sentiment classification framework grounded in FinBERT, augmented with multiple normalization strategies to aggregate sentence-level outputs into article-level and daily sentiment indices. Through a large-scale human-annotated evaluation, the work provides the first systematic validation of the alignment between automated sentiment metrics and human assessments. Results demonstrate that the proposed method significantly outperforms conventional lexicon-based techniques across correlation, regression, and classification tasks, achieving superior precision in distinguishing positive, neutral, and negative sentiment while exhibiting strong consistency with human consensus.

Abstract

Measuring sentiment from financial news is a central task in economics and finance, yet most existing indicators rely on dictionary-based approaches that infer sentiment from word counts and only partially capture context, negation, and semantic structure. This paper proposes a framework for constructing daily news mood indices using transformer-based language models and evaluates whether they better represent sentiment than dictionary-based alternatives. Using 143,755 financial news articles from Factiva, we classify sentiment at the sentence level with FinBERT and aggregate these predictions into article-level and daily sentiment measures through alternative normalization schemes. We compare the resulting indices with benchmark measures based on Shapiro et al., 2022 and Barbaglia et al., 2025. A central contribution is the validation of alternative sentiment measures against human judgments. We conducted an incentivized annotation exercise in which 444 participants evaluated a validation subsample of 588 financial news articles. Consensus ratings from independent human evaluations serve as an external benchmark for assessing the quality of automated sentiment measures. Across correlation, regression, and classification exercises, transformer-based measures show stronger agreement with human judgments than vocabulary-based alternatives and perform substantially better in distinguishing positive, neutral, and negative articles. Overall, the results suggest that incorporating contextual information through transformer-based language models produces sentiment measures that more closely reflect human assessments of financial news.

Problem

• sentiment analysis
• financial news
• dictionary-based methods
• contextual understanding
• human judgment

Innovation

• Transformer-based language models
• FinBERT
• financial news sentiment
• human-validated sentiment
• context-aware sentiment analysis