Comparing social media comments about children's vaccines with national survey findings

Background—For health communication campaigns to be effective, it is important to understand the targe populations' beliefs, attitudes, and perceived norms. This information is typically collected using surveys or focus groups, which can be expensive. Social media may provide a proxy for survey findings on beliefs, attitudes, and norms for certain populations. If so, cost-efficient methods of analyzing social media posts can help inform communication initiatives without imposing burden on the public.

Objective—To compare findings from analysis of social media content with findings from a national survey.

Methods--We used Natural Language Processing to analyze comments about childhood vaccinations from a social media forum popular with new and expecting mothers.  We developed a sentiment analysis classification tool by manually coding 300 postings on the forum and training a computer learning algorithm to classify about 33,000 postings into positive, negative, and neutral sentiments. We conducted topic modeling using Latent Dirichlet Allocation on the social media postings to determine a number of distinct topics.

Results--The sentiment analysis classifier was not accurate, perhaps because attitudes toward childhood vaccination is complicated—parents may  agree with the vaccine schedule but cringe when their children are stuck with needles. Topic modeling proved more useful; 25 topics were produced, each representing a profile of parents’ positions on vaccination. Some topics found by modeling were absent from the survey, which indicates that this analysis may be useful for checking content validity of survey instruments.  

Discussion—Our next step will compare the findings from the social media site with responses to the NIS. Analyses will assess the extent to which the data from social media and surveys correlate, and will provide guidelines to communication professionals and researchers on opportunities and limitations for using social media content analysis to help prepare messages for public health campaigns.