Complexities Associated with Sentiment Mining Foursquare Check-ins

Abstract

Broad adoption of smartphones has increased the number of messages generated while people are going about their daily lives. Many of these messages are related to the location where that message is generated. Being able to infer a person's sentiment toward a given location and aggregate that would be a boon to marketers and market researchers. I examined a set of nearly 30,000 messages from Foursquare that had been cross-posted to Twitter to find the best way to assign sentiment from each message to its associated location. After relatively brief review of the data, it was clear calculating sentiment and assigning that sentiment to the location associated with the message would give an inaccurate relative sentiment score to many locations. I attempted to use naive Bayes to determine if a message was more or less likely to be associated with an individual's present location. A simple application of naive Bayes had a reasonable precision but discarded too many messages. More advanced machine learning methods should be explored to differentiate between messages a user makes that are, and are not located to where they are when they sent said message.