Impact of Social Networks on Buying Behavior: Predicting the Success of Pittsburgh Businesses Through Analysis of Yelp Social Networks

Abstract

In this paper, we analyze data collected from Yelp to understand the importance of the social networks created between Yelp reviewers, and the impact it has on the businesses they patronize. We then look at the shape of the social network generated by reviews and identify differences between the behavior of “elite” users vs. non-elite users. We then view the trends in network formation for particular businesses over time, and use features of the network of the business early in its development in order to predict its future success. We construct linear regression and random forest models that solely use features derived from review data, as well as models that are built using a combination of review and social network features. We see that the additional network features are statistically signifant, and help reduce the root mean squared error of our models by a significant percentage. Ultimately, using our network features of reviewers from the first three months of business, we can predict the number of reviewers for a business within its first year with an error of less than 2.5 reviews, an error of 8.5 over two year, and 13 over three.