Author: 
Tao Chen
Mark Dredze
Publication Date
April 3, 2018
Affiliation: 

Center for Language and Speech Processing, Johns Hopkins University

"Despite widespread acceptance that images are crucial in public health messaging, there is no visual theory that systematically guides the design and use of images, and characteristics of effective images in health communication remain unclear..."

Considering that visual imagery plays a key role in health communication, this study sought to understand how images are used in vaccine-related tweets and to provide guidance with respect to the characteristics of vaccine-related images that correlate with a higher likelihood of being retweeted. As the researchers explain, both vaccine skeptics and supporters have a large presence on social media in general and Twitter in particular, and they use these platforms to advocate for their positions. They often use images, which are not only effective tools for eliciting emotional reactions but also for conveying statistics and data in support of a position (e.g., in the form of an infographic). In fact, as reported here, images have been used in vaccine tweets (1,137,172/6,288,653; 18.08% of vaccine tweets contain an image) but have not been studied. To that end, these researchers explore two questions: (i) What are the characteristics of these images? and (ii) what are the kinds of traits that make them engaging?

The researchers collected more than one million vaccine image messages from Twitter (from November 11 2014 to August 8 2016) and characterised various properties of these images using automated image analytics. They fit a logistic regression model to predict whether or not a vaccine image tweet was retweeted, thus identifying characteristics that correlate with a higher likelihood of being shared. For comparison, they built similar models for the sharing of vaccine news on Facebook and for general image tweets.

As with the previous work on general image tweets, an image makes a vaccine tweet more likely to be shared (72,906/237,478;30.70% of image tweets were retweeted) than their text-only counterparts (483,815/3,162,184; 15.30%). This is a possible motivation for users to attach images to tweets. The large number of vaccine images from Twitter that are duplicates (125,916/237,478; 53.02%) or found on other websites (442/500,88.4%) suggests that most vaccine images are not user created (e.g., a photo taken by a user); instead, they are selected from other sources by the user to help promote their vaccination message. Also, the much higher proportion of vaccine image tweets with external URLs (653,874/1,137,172; 57.50%) compared with general image tweets (22.7%) suggests that images play an important role in vaccine-related messaging. Furthermore, many of the vaccine images contain their own information beyond a visual supporting of the message in the tweet's text. Nearly 40% of images have embedded text, and embedded text is informative to interpret the overall message of the tweet. These images include screenshots, infographics, charts, and figures.

Focusing on the visual content, the two most recurring objects in vaccine images are syringes and people. The visual content is also highly correlated with a tweet's textual topics. As such, the purpose of attaching an image to a tweet is to make it more attractive and convey the topics of the tweet. Vaccine tweets with images were twice as likely to be shared as nonimage tweets, which follows the trend of general tweets. The logistic regression identified the author as one of the most important factors for determining whether an image tweet would be shared, the same as in general tweets. Sentiment features, extracted from text and image, are also predictive of sharing behaviour. Positive or negative sentiment vaccine image tweets are more likely to be retweeted than neutral tweets. Although previous work found that images with faces have a higher user engagement, the present study found that vaccine image tweets containing a face were equally likely to be retweeted as those without a face (25.5% with vs 25.7% without).

The researchers found that differing behaviours of features between vaccine tweets and vaccine news. For instance, a smiling face increased sharing for vaccine tweets, but not news, whereas pictures of landscape and nature contribute positively for news sharing but negatively for tweets. This suggests that different communication patterns exist in the two domains (tweet and news), or there could be a difference in how people decide to share content on the two social media platforms (Twitter and Facebook).

The researchers suggest that vaccine-related communication strategies could benefit from their analyses. Images boost the reach of a vaccine message. The retweet predictive model outlined in the paper could be used as a tool to assess the effectiveness of designed visual vaccine messages. From that model, the researchers also identify a few key factors that correlate with the retweeting of vaccine tweets. Although they have not established a causal relation, these factors could still guide message design. Finally, the study "demonstrates an effective methodology for image analysis studies. We found that Twitter is a productive platform for studying visual communication issues surrounding vaccines. This is an important finding because Twitter makes it relatively easy to collect large quantities of image data via the public Twitter API [application programming interface]....Unlike prior work that relied on human analysis of images, we used fully automated analytics to conduct a comprehensive analysis over a large dataset. Such techniques can be applied to analyze vaccine images from other sources and health-related images in general. To enable future studies, we have released the labeled datasets that were used to build our vaccine relevance and sentiment classifiers..."

According to the researchers, several large social media platforms, including Instagram and Facebook, in which images are prevalent, have not been examined for vaccine content. Because effective messaging strategies need to be tailored for each platform, evidence of vaccine image effectiveness on these platforms could be pursued and provided as part of future research studies. In addition, questions related to understanding images beyond the analytics presented in this paper remain; they could include: What images are most effective for different campaigns? How do images tie into existing narratives around vaccination? How are intended populations of vaccination campaigns reflected in images? These questions could be applied broadly to public health awareness campaigns.

Source: 

Journal of Medical Internet Research Vol 20, No 4 (2018): April. http://www.jmir.org/2018/4. Image credit: Pixabay; Copyright: Angelo Esslinger. License: Public Domain (CC0)