A Computational Framework Using NLP and LLMs for Media Bias Detection in Geopolitical News Reporting
Author
Hasan, Mohammad Rakib
Term
4. semester
Publication year
2026
Submitted on
2026-06-01
Pages
60
Abstract
This thesis explores how AI-based language systems can help detect and assess media bias when Bangladeshi and international outlets cover the same geopolitical events. Media bias is understood here as systematic differences in how stories are selected, framed, and worded. The study examines coverage of the Russia–Ukraine war, the Iran–Israel conflict, and tensions in the Taiwan Strait to see how political narratives differ across regions and how automated systems interpret those differences. To do this, the project builds a computational framework that brings together automated news collection (web scraping), Natural Language Processing (NLP, methods for analyzing text), multilingual handling, and analysis with large language models (LLMs). Articles from the Daily Star, New Age, BBC, and The Guardian were collected and preprocessed by removing duplicates, filtering for relevant topics, and standardizing the text. The analytical stage combined lexical analysis (patterns in words and phrases), sentiment evaluation (emotional tone), and LLM-based comparative interpretation. Rather than relying only on traditional classifiers, the system generated contextual summaries and side-by-side comparisons that highlighted differences in framing, tone, geopolitical emphasis, and how key actors were portrayed across sources. The framework consistently found meaningful contrasts between Bangladeshi and international reporting. In the selected material, Bangladeshi coverage placed stronger emphasis on humanitarian consequences, while international coverage focused more on diplomacy, geopolitical strategy, and wider global implications. Several forms of bias—framing, selection, and linguistic—were observed in both media environments. Overall, the results show that NLP and LLM-based systems can support scalable and interpretable comparative media analysis, offering both a methodological framework for computational bias detection and a practical demonstration for media studies and Business Data Science.
Denne afhandling undersøger, hvordan AI-baserede sprogsystemer kan hjælpe med at opdage og vurdere mediebias, når bangladeshiske og internationale medier dækker de samme geopolitiske begivenheder. Mediebias forstås her som systematiske forskelle i, hvilke historier der udvælges, hvordan de vinkles, og hvilke ord der bruges. Studiet ser på dækningen af krigen mellem Rusland og Ukraine, konflikten mellem Iran og Israel samt spændingerne i Taiwanstrædet for at afdække, hvordan politiske fortællinger varierer på tværs af regioner, og hvordan automatiske systemer tolker disse forskelle. Til formålet udvikles en beregningsmæssig ramme, der kombinerer automatisk nyhedsindsamling (webscraping), Natural Language Processing (NLP, metoder til tekstanalyse), flersproglig håndtering og analyse med large language models (LLM’er). Artikler fra Daily Star, New Age, BBC og The Guardian blev indsamlet og forbehandlet ved at fjerne dubletter, filtrere for relevante emner og standardisere teksten. I den analytiske fase blev leksikalsk analyse (mønstre i ord og fraser), sentimentanalyse (følelsestonen) og LLM-baseret komparativ fortolkning kombineret. Systemet genererede kontekstuelle resumeer og side-om-side-sammenligninger, der – i stedet for kun at bruge traditionelle klassifikatorer – synliggjorde forskelle i vinkling, tone, geopolitisk fokus og portrætteringen af centrale aktører på tværs af kilder. Rammen fandt gennemgående meningsfulde kontraster mellem bangladeshisk og international dækning. I det udvalgte materiale lagde bangladeshiske medier stærkere vægt på humanitære konsekvenser, mens internationale medier i højere grad fremhævede diplomati, geopolitisk strategi og bredere globale implikationer. Flere former for bias – framing-, selektions- og sproglig bias – blev observeret i begge mediemiljøer. Samlet viser resultaterne, at NLP- og LLM-baserede systemer kan understøtte skalerbar og gennemskuelig komparativ medieanalyse og bidrager med både en metodisk ramme for computermæssig biasdetektion og et praktisk eksempel til mediestudier og Business Data Science.
[This apstract has been rewritten with the help of AI based on the project's original abstract]
Keywords