Analyzing Sentiment Polarity Reduction in News Presentation through Contextual Perturbation and Large Language Models
Alapan Kuila, Somnath Jena, Sudeshna Sarkar, Partha Pratim Chakrabarti
TL;DR
This paper tackles the problem of sentiment polarity manipulation in news text by introducing two complementary strategies: adversarial-contextual perturbations and ChatGPT-based prompt engineering, both designed to neutralize sentiment toward specific news aspects while preserving core semantics. The authors define a formal objective to maximize the neutral probability $P_s(S',a_t)$ for aspect $a_t$ and target sentiment $s$, under constraints that preserve meaning and fluency. They implement three transformation operations (replace, insert, delete) with a beam search guided by an NLI-based sentiment classifier and evaluate on a GDELT-derived, aspect-annotated corpus, complemented by human judgments. Findings indicate that replacement-based perturbations under proper constraints achieve strong neutrality with high entailment and text preservation, while ChatGPT offers higher semantic fidelity but lower neutrality, suggesting a trade-off between targeted perturbation precision and broader language-model capabilities. The work contributes a parallel corpus of original and neutral sentences and demonstrates a feasible route toward more objective news reporting with potential impact on media bias mitigation and newsroom practices.
Abstract
In today's media landscape, where news outlets play a pivotal role in shaping public opinion, it is imperative to address the issue of sentiment manipulation within news text. News writers often inject their own biases and emotional language, which can distort the objectivity of reporting. This paper introduces a novel approach to tackle this problem by reducing the polarity of latent sentiments in news content. Drawing inspiration from adversarial attack-based sentence perturbation techniques and a prompt based method using ChatGPT, we employ transformation constraints to modify sentences while preserving their core semantics. Using three perturbation methods: replacement, insertion, and deletion coupled with a context-aware masked language model, we aim to maximize the desired sentiment score for targeted news aspects through a beam search algorithm. Our experiments and human evaluations demonstrate the effectiveness of these two models in achieving reduced sentiment polarity with minimal modifications while maintaining textual similarity, fluency, and grammatical correctness. Comparative analysis confirms the competitive performance of the adversarial attack based perturbation methods and prompt-based methods, offering a promising solution to foster more objective news reporting and combat emotional language bias in the media.