Where Not to Eat? Improving Public Policy by Predicting Hygiene Inspections Using Online Reviews

Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing, pages 1443–1448, Seattle, Washington, USA, 18-21 October 2013. c⃝2013 Association for Computational Linguistics Where Not to Eat? Improving Public Policy by Predicting Hygiene Inspections Using Online Reviews Jun Seok Kang† Polina Kuznetsova† †Department of Computer Science Stony Brook University Stony Brook, NY 11794-4400 {junkang,pkuznetsova,ychoi} @cs.stonybrook.edu Michael Luca‡ Yejin Choi† ‡Harvard Business School Soldiers Field Road Boston, MA 02163 mluca@hbs.edu Abstract This paper offers an approach for governments to harness the information contained in social media in order to make public inspections and disclosure more efficient. As a case study, we turn to restaurant hygiene inspections – which are done for restaurants throughout the United States and in most of the world and are a fre- quently cited example of public inspections and disclosure. We present the first empiri- cal study that shows the viability of statistical models that learn the mapping between tex- tual signals in restaurant reviews and the hy- giene inspection records from the Department of Public Health. The learned model achieves over 82% accuracy in discriminating severe offenders from places with no violation, and provides insights into salient cues in reviews that are indicative of the restaurant’s sanitary conditions. Our study suggests that public disclosure policy can be improved by mining public opinions from social media to target in- spections and to provide alternative forms of disclosure to customers. 1 Introduction Public health inspection records help customers to be wary of restaurants that have violated health codes. In some counties and cities, e.g., LA, NYC, it is required for restaurants to post their inspec- tion grades at their premises, which have shown to affect the revenue of the business substantially (e.g., Jin and Leslie (2005), Henson et al. (2006)), thereby motivating restaurants to improve their sani- tary practice. Other studies have reported correlation between the frequency of unannounced inspections per year, and the average violation scores, confirm- ing the regulatory role of inspections in improving the hygiene quality of the restaurants and decreasing food-borne illness risks (e.g., Jin and Leslie (2003), Jin and Leslie (2009), Filion and Powell (2009), NYC-DoHMH (2012)). However, one practical challenge in the current inspection system is that the department of health has only limited resources to dispatch inspectors, leaving out a large number of restaurants with un- known hygiene grades. We postulate that online re- views written by the very citizens who have visited those restaurants can serve as a proxy for predicting the likely outcome of the health inspection of any given restaurant. Such a prediction model can com- plement the current inspection system by enlight- ening the department of health to make a more in- formed decision when allocating inspectors, and by guiding customers when choosing restaurants. Our work shares the spirit of recently emerging studies that explores social media analysis for pub- lic health surveillance, in particular, monitoring in- fluenza or food-poisoning outbreaks from micro- blogs (e.g., Aramaki et al. (2011), Sadilek et al. (2012b), Sadilek et …

Content truncated to 500 words