Finding trends and statistical patterns in name mentions in news

TL;DR

This study analyzes trends and statistical patterns in news mentions of individual names over time, revealing stable power-law distributions and dual scaling behaviors, and models these patterns using social network structures.

Contribution

It introduces a novel method for extracting and analyzing name mentions in news, demonstrating consistent power-law distributions and dual scaling, and models social influence using Barabasi-Albert networks.

Findings

Mentions of names fluctuate strongly over time.

Rank-frequency distributions follow power-laws similar to Zipf's law.

Names are archived differently from regular words, showing dual scaling behaviors.

Abstract

We extract the individual names of persons mentioned in news reports from a Philippine-based daily in the English language from 2010-2012. Names are extracted using a learning algorithm that filters adjacent capitalized words and runs it through a database of non-names grown through training. The number of mentions of individual names shows strong temporal fluctuations, indicative of the nature of "hot" trends and issues in society. Despite these strong variations, however, we observe stable rank-frequency distributions across different years in the form of power-laws with scaling exponents \alpha = 0.7, reminiscent of the Zipf's law observed in lexical (i.e. non-name) words. Additionally, we observe that the adjusted frequency for each rank, or the frequency divided by the number of unique names having the same rank, shows a distribution with dual scaling behavior, with the higher-ranked names preserving the \alpha exponent and the lower-ranked ones showing a power-law exponent \alpha' = 2.9. We reproduced the results using a model wherein the names are taken from a Barabasi-Albert network representing the social structure of the system. These results suggest that names, which represent individuals in the society, are archived differently from regular words.