Mining the Twitter‐Sphere for Consumer Attitudes Towards Dairy

Historically, the study of consumer attitudes and beliefs about food have relied heavily on self‐reported data using survey and interviewing methodologies. Opinion mining of social media data may be an effective method for collecting and evaluating real‐time attitudes regarding food choice and dietary intake patterns. This study was a post‐hoc analysis of geographical tweets across the continental United States to assess attitudes toward dairy products. Between the years August, 2013 – June, 2015, 232628 geo‐located tweets were collected from the social media application, Twitter, using the following keywords: dairy, milk, cheese, yogurt and kefir. Natural language processing techniques including, topic extraction and sentiment analysis were used to analyze the data. Latent Dirichlet Allocation (LDA) models were used to establish contextual probabilities for discussion topics and account for different contexts by state and geographical region (South, North East, West and Mid‐West). Sentiment analysis of each tweet was completed using the Stanford CoreNLP, SentiStrength and SentiWordNet toolkits and sentiment values were combined to assign an overall sentiment score. Negative and positive sentiment scores were assigned to each tweet based on a scale from −1 to 1. Principal components analysis was used to assess the significance of tweet topics for all states. Five major topics were identified from these analyses: milk, dairy queen, cheese, chocolate milk and yogurt. Tweet topics relating to milk and cheese were the most common topics throughout the United States with even distributions across the four country regions. Cheese‐related topics were twice as common as chocolate milk tweets across all states; this trend was also apparent in the analysis by region. Yogurt was mentioned more frequently in the West while Dairy Queen was more commonly mentioned in Mid‐West and Southern regions. Tweet topics from the District of Columbia differed from other states with significantly more mentions of chocolate, cheese and yogurt. Wyoming and Montana had the lowest overall significance values for major identified topics. Sentiment analyses revealed that cheese‐related topics received mixed attitudes while chocolate milk was positively discussed across the United States. Dairy Queen mentions were very positive but discussion of yogurt was slightly more negative in the East when compared to the West. Overall, Twitter discussions of dairy foods were positive but regional differences were noted for yogurt. Social media mining provides a novel method of studying food attitudes and behaviors in real time. Such analyses may be beneficial for public health and policy initiatives and allow health/nutrition educators to more adequately address region‐ and state‐specific feelings about foods and beverages. Support or Funding Information N/A