Cati Gene Brown-johnson*, UCSF, San Francisco, United States Ashley Sanders-jackson, UCSF, San Francisco, United States Judith J Prochaska, UCSF, San Francisco, United States Track: Research Presentation Topic: Web 2.0 approaches for behaviour change, public health and biosurveillance Presentation Type: Poster presentation Submission Type: Single Presentation Last modified: 2012-09-12

If you are the presenter of this abstract (or if you cite this abstract in a talk or on a poster), please show the QR code in your slide or poster (QR code contains this URL).

Background: Understanding information contagious on Twitter (i.e., what is passed on from one user to another, or “retweeted”) may help to inform more effective online health interventions.

Objective: The current study analyzed semantic network structure and attributes of retweeted and non-retweeted Twitter messages (“tweets”) following two health-related media events: 1) Miley Cyrus’ fans discovering she smoked cigarettes and 2) an announcement that President Obama quit smoking.

Methods: Tweet data were collected using search terms over a multiple day period. Final counts were: 48 unique retweets and 263 non-retweeted messages for President Obama data; 274 unique retweets and 6660 non-retweeted messages for Miley Cyrus data.

This study operationalized information contagion via a semantic network methodology treating words as network nodes. Two types of semantic network analyses were completed to examine differences between contagious information (retweets) and other (non-retweeted) messages: 1) a network visualization of terms and 2) an exponential random graph model of underlying network structure. Additionally, the use of “#” to associate a comment with similar comments (“hashtag”), online links (“URLs”), and word frequency were examined.

Results:
The underlying network structure of messages that get retweeted was fundamentally different from messages that did not. For example, in the Miley Cyrus dataset retweets had significantly more triads (connections between three words) compared to other messages. There was substantial overlap in high frequency terms in both retweeted and non-retweeted messages. This overlap decreased as word frequency decreased. Lower frequency terms tended to fall in either the retweet or non-retweeted conversation. A striking example of this in the President Obama dataset involved the n-word and “lips”, which were moderately-used terms that only appeared in the retweeted network.

Miley Cyrus retweets contained more hashtags and URLs than non-retweeted messages; whereas, in the President Obama dataset, retweets contained less hashtags and URLs than non-retweeted messages. Regardless, for both datasets, hashtags and URLs were more central in retweeted networks; that is, they were positioned more centrally within the semantic networks and they linked with larger amounts of information even though they occurred less frequently that other categories of information (e.g. political terms in the President Obama dataset).

In the Miley Cyrus dataset, the two most frequent hashtags (#welovemileynomatterwhat and #mileypleasestopsmoking) were the same for both the retweet and non-retweet messages, reflecting the similarity in topic for messages that were retweeted and those that were not.

Conclusion: Though the Miley Cyrus and President Obama cases contained very different content and most likely involved very different participants, there were similarities across the cases. Regardless of frequency, hashtags and website links are central to networks. Tying into pertinent hashtags and linking to more traditional news coverage may make health information more “sticky”. An open question is whether terminology used by out-group entities, like public health practitioners, will be received and passed-on. Though there was some participation of public health practitioners in the President Obama data, this did not occur at all in the Miley Cyrus dataset.