Identifying Market Structure: A Deep Network Representation Learning of Social Engagement

Abstract

With rapid technological developments, product-market boundaries have become more dynamic. Consequently, competition for products and services is emerging outside the product-market boundaries traditionally defined based on Standard Industrial Classification and North American Industry Classification System codes. Identifying these fluid product-market boundaries is critical for firms not only to compete effectively within a market but also to identify lurking threats and latent opportunities outside market boundaries. Newly available big data on social media engagement presents such an opportunity. The authors propose a deep network representation learning framework to capture latent relationships among thousands of brands and across many categories, using millions of social media users’ brand engagement data. They build a brand–user network and then compress the network into a lower-dimensional space using a deep autoencoder technique. The authors evaluate this approach quantitatively and qualitatively and visually display the market structure using the learned representations of brands. They validate the learned brand relationships using multiple external data sources. They also illustrate how this method can capture the dynamic changes of product-market boundaries using two well-known events—the acquisition of Whole Foods by Amazon and the introduction of the Model 3 by Tesla—and how managers can use the insights that emerge from this analysis.