Zika discourse in the Americas: A multilingual topic analysis of Twitter

Abstract

This work examines Twitter discussion surrounding the 2015 outbreak of Zika, a virus that is most often mild but has been associated with serious birth defects and neurological syndromes. We introduce and analyze a collection of 3.9 million tweets mentioning Zika geolocated to North and South America, where the virus is most prevalent. Using a multilingual topic model, we automatically identify and extract the key topics of discussion across the dataset in English, Spanish, and Portuguese. We examine the variation in Twitter activity across time and location, finding that rises in activity tend to follow to major events, and geographic rates of Zika-related discussion are moderately correlated with Zika incidence (ρ = .398).

Fig 1. Crosslingual consistency versus number of translation pairs.

Learning curves showing how crosslingual consistency of topics, measured by MTA, varies with the number of pairs of translated documents in the corpus.

Fig 2. Volume of Zika tweets across time and place.

Top: The adjusted volume of Zika-related tweets in five geographic regions per week. The country flags below the line plots indicate the time window in which each country reported its first Zika case. Bottom: Tweet volume in each country during six time windows spanning our data collection. Darker shading indicates higher volume; the color scale is on a log₂ scale.

Fig 3. Topic prevalence by location.

Geospatial distribution of the topics from Table 3. Darker shading indicates a higher average probability of the topic in that location.

Fig 4. Topic prevalence by time.

The volume and distribution of the topics from Table 3 per week. The adjusted counts of each topic are shown on top; dashed lines are added for readability and do not have a special meaning. The adjusted counts are normalized to sum to 1 on the bottom; the shaded gray area represents the proportion of the 36 other topics outside of the 14 topics in Table 3.