Can User Generated Content Predict Restaurant Survival: Deep Learning of Yelp Photos and Reviews

46 Pages Posted: 31 Jan 2018 Last revised: 28 Mar 2018

See all articles by Mengxia Zhang

Mengxia Zhang

University of Southern California

Lan Luo

University of Southern California

Date Written: March 2018

Abstract

With widespread usage of smartphones, 3 billion photos are shared on the Internet daily. Nevertheless, very few papers have examined whether and how the prosperity of a business may be affected by consumer posted photos. We use deep learning methods to analyze 795,175 photos and 1,015,825 reviews posted on Yelp from 2004 to 2015 on 17,796 restaurants. Tracking the survival of these restaurants during this time period, we find that both volume and valence of photos are strong predictors of restaurant survival. Nevertheless, when it comes to reviews, only valence (not volume) matters. Interestingly, even after controlling for the content, star rating, and length of reviews, consumer sentiment extracted from review text is still strongly associated with the survival of restaurants. We also find that chain restaurants and restaurants from larger categories have greater chance to survive. Ceteris paribus, price levels do not appear to impact restaurant survival. Our research is among the earliest attempts to introduce both photo- and text- based deep learning in marketing. We are also the first to compare and contrast managerial impacts of consumer posted photos vs. reviews. To our knowledge, this is also the first large-scale empirical research on restaurant survival.

Keywords: marketing analytics, consumer posted photo, image recognition, computer vision, image valence, review sentiment, text mining, deep learning, convolutional neural network, restaurant survival, yelp

JEL Classification: A10, C10, C13, C41, C45, C50, C53, C55, D00, E00, E02, L10, L25, M10, M21, M31

Suggested Citation

Zhang, Mengxia and Luo, Lan, Can User Generated Content Predict Restaurant Survival: Deep Learning of Yelp Photos and Reviews (March 2018). Available at SSRN: https://ssrn.com/abstract=3108288 or http://dx.doi.org/10.2139/ssrn.3108288

Mengxia Zhang

University of Southern California ( email )

2250 Alcazar Street
Los Angeles, CA 90089
United States

Lan Luo (Contact Author)

University of Southern California ( email )

Hoffman Hall 319
Los Angeles, CA 90089-1427
United States

Register to save articles to
your library

Register

Paper statistics

Downloads
192
rank
151,082
Abstract Views
572
PlumX Metrics