(Im)balance in the Representation of News? An Extensive Study on a Decade Long Dataset from India

Souvic Chakraborty, Pawan Goyal, Animesh Mukherjee

Introduction

Media bias and political imbalance are central themes in contemporary democratic discourse. While often discussed qualitatively, quantifying these phenomena scientifically presents significant challenges. Most computational research in this field has been limited to short-term analyses or focused predominantly on Western contexts. This study, conducted by researchers at IIT Kharagpur, addresses this gap by providing a comprehensive, decade-long quantitative analysis of media imbalance within the Indian news landscape. By examining over 3.8 million articles from 2010 to 2018, the research identifies systematic patterns in how major English-language newspapers represent political parties, geographic regions, and national priorities.

Methodology

The study moves beyond simple keyword frequency to employ a multi-dimensional suite of NLP and statistical techniques. The researchers adapted the Lacy metric to calculate imbalance across four primary lenses:

  • Coverage: The volume of headlines and body text dedicated to specific political entities.
  • Point of View (PoV): The frequency of narrative verbs (e.g., "claims," "states") attributed to party members.
  • Tonality: Sentiment analysis using VADER and subjectivity scores via TextBlob.
  • Superlatives: The use of comparative and superlative adjectives as proxies for descriptive bias.

To identify systemic similarities between outlets, the team utilized Dynamic Time Warping (DTW) and hierarchical clustering. Furthermore, Word Embedding Association Tests (WEAT) were used to measure the "popularity" of political parties by associating them with pleasant or unpleasant word sets. Finally, the RoBERTa language model was employed for cloze tasks to probe shifting national priorities reflected in media language.

Data Corpus

The analysis is based on a massive dataset of 3.86 million political articles from three leading Indian English-language outlets:

  1. The Times of India (TOI): Approximately 1.9 million articles.
  2. The Hindu: Approximately 1.03 million articles.
  3. India Today (IT): Approximately 0.92 million articles.

The data includes headlines, publication dates, and full-text content, specifically filtered for keywords related to the Bharatiya Janata Party (BJP) and the Indian National Congress (INC).

Analysis and Insights

A key finding of the study is the existence of relatively static ideological leanings among media houses. For example, The Hindu consistently exhibited a 5–10% shift in coverage toward the Congress party compared to its peers.

Clustering analysis further revealed distinct behavioral patterns. The Times of India and India Today frequently clustered together across various imbalance metrics, suggesting a shared reporting style or priority set. In contrast, The Hindu often emerged as an outlier, maintaining a distinct reporting profile throughout the decade.

Dendrogram illustrating the hierarchical clustering of newspapers, showing the close relationship between India Today and TOI compared to The Hindu.

The study also tracked the "popularity crossover" around 2014, where WEAT scores indicated that the BJP's association with positive descriptors in news discourse surpassed that of the Congress—a trend that has persisted.

Results

The aggregate results show that TOI exhibited the highest absolute imbalance in headlines and Point of View. Conversely, The Hindu showed the highest imbalance in content volume and positive sentiment. Interestingly, The Hindu often displayed high positive and high negative sentiment for the same party simultaneously, suggesting that imbalance can manifest as intense, polarized discussion rather than simple one-sidedness.

Hierarchical clustering results demonstrating consistent reporting patterns between TOI and India Today, with The Hindu as an outlier.

Geographically, the study confirms a significant skew toward urban centers like Delhi and Mumbai, which account for 60–70% of city-level mentions. Regions such as the North-East and Jammu & Kashmir remain significantly underrepresented, highlighting persistent "news deserts."

Conclusion and Future Work

While providing a robust framework for quantifying media imbalance, the study acknowledges limitations, including its focus on the two-party system and English-language media. Future research aims to extend this framework to multi-lingual settings to capture the nuances of regional media and explore the evolution of religious and community-wise polarization in Indian society.

Summary updated

BibTeX
@inproceedings{chakraborty2022im,
  author = {Souvic Chakraborty and Pawan Goyal and Animesh Mukherjee},
  title = {{(Im)balance in the Representation of News? An Extensive Study on a Decade Long Dataset from India}},
  booktitle = {SocInfo},
  year = {2022},
  url = {https://openreview.net/forum?id=jIhKWfCC8R},
}
Figure