[Paper Review] What Gets Media Attention and How Media Attention Evolves Over Time - Large-scale Empirical Evidence from 196 Countries

Jisun An, Haewoon Kwak|arXiv (Cornell University)|Apr 5, 2017

Complex Network Analysis Techniques10 references3 citations

TL;DR

This study analyzes large-scale global news data from 196 countries (March–October 2016) to identify what topics receive media attention and how attention evolves over time. Using dynamic time warping and K-means++ clustering, it identifies four distinct temporal patterns—constant coverage, multi-day peaks, sharp single-day peaks, and frequent sharp peaks—revealing that most topics have short attention spans, with only 2.22% showing sustained media focus.

ABSTRACT

It is known that news topics, covered more frequently and over longer periods of time, are considered to be important to the public. Hence, what gets media attention and how media attention evolves over time has been studied for decades in communication study. However, previous studies are confined to a few countries or a few topics, mainly due to lack of longitudinal global data. In this work, we use a large-scale news data compiled from 196 countries to provide empirical analyses of media attention dynamics.

Motivation & Objective

• To identify global patterns in media attention across 196 countries using large-scale longitudinal news data.
• To investigate whether media attention dynamics vary by topic type (e.g., country, person, event) and region.
• To classify temporal evolution patterns of media attention and assess their prevalence and characteristics.
• To provide empirical validation of the alarm/patrol media model at scale, beyond prior case studies.
• To quantify the attention span of news topics and identify factors influencing sustained or transient coverage.

Proposed method

• Collected and processed a large-scale news dataset from 196 countries via Unfiltered News, using machine translation to index multilingual content.
• Extracted topic metadata and filtered for fine-grained types (e.g., Domestic Country, Foreign City), removing hierarchical redundancies.
• Computed daily topic attention proportions per country and averaged across countries to derive global attention profiles.
• Applied dynamic time warping (DTW) to measure similarity between time-series of media attention, enabling clustering of temporal patterns.
• Used K-means++ clustering with elbow method to determine four distinct clusters of media attention evolution patterns.
• Performed linear normalization of time-series to reduce bias in DTW distance computation before clustering.

Experimental results

Research questions

• RQ1 What types of topics receive the most media attention globally, and is there a consistent global order of attention?
• RQ2 What is the typical attention span of news topics across countries, and which topics sustain long-term media coverage?
• RQ3 Do media attention patterns follow distinct temporal shapes, and if so, how do they vary by topic type?
• RQ4 To what extent do regional or topic-specific factors influence the duration and intensity of media coverage?
• RQ5 How do the findings align with the alarm/patrol media model in real-world, large-scale data?

Key findings

• Domestic Country is the top-ranked topic type in media attention, consistently receiving the highest proportion of coverage across all months.
• The average rank correlation of topic types across months (excluding June) is 0.96, indicating high stability in media attention order except during global events.
• 76.32% of media attention time-series exhibit a sharp peak and rapid decline (Cluster C3), confirming a dominant short attention span in news media.
• Only 2.22% of topics show constant media attention (Cluster C1), primarily including Domestic Countries and key figures like Donald Trump or Hilary Clinton.
• Foreign Country topics show the highest proportion (22.4%) in the multi-day peak pattern (Cluster C2), aligning with selective foreign news coverage.
• Domestic Cities have a relatively high proportion (12.1%) in the constant coverage cluster (C1), but most follow the short-lived peak pattern (C3), indicating attention inequality in local media.