When looking for a new place to eat, diners often turn to online restaurant reviews for help. There is a less obvious, but just as valuable, insight one can extrapolate than simply finding the best bagel or a cute date-spot night. Shawn Mankad, Samuel Curtis Johnson Graduate School of Management, has developed a way to analyze online restaurant reviews to effectively identify a restaurant’s hygiene violations.
Mankad is a data scientist and statistician by training. He’s interested in using tools and methods from his field to solve business and policy issues.
Restaurant Hygiene, a Public Concern
Mankad started analyzing online reviews primarily because the data was available—a low barrier to entry for conducting research. There is, however, a whole industry and thousands of papers written on analyzing online reviews. “We were trying to do something a little different,” he says. “Can we use an online review to answer a question that has public health policy consequences and have a little more impact?”
When it comes to restaurant hygiene, consumers and governments depend on health inspection grades. Health inspections, however, take place infrequently. Mankad calls this a classical problem in economics. It is impossible to inspect restaurants all of the time, yet hygiene is a constant, pressing concern.
Using machine learning techniques, Mankad created a hygiene dictionary and analysis tool—combing through online reviews for keywords and phrases, such as hair in food, glass stained, and filthy. He compared the frequency of these words to data from restaurant hygiene inspections in New York City between 2010 and 2016.
“We find that our measure is predictive of how the restaurant will score,” says Mankad. “And the second thing we find is evidence for moral hazard. After a restaurant gets an A rating, they're not going to be inspected for a year; and often after the A grade, they let the foot off the gas.”
The model gets smarter, continuing to learn indicative keywords and phrases. A tool like this could distill the wealth of crowd-sourced information available online and offer current policymakers a helpful solution. “It’s possible that cities and local governments could use online reviews to inform what they’re doing,” says Mankad. “They could, for example, schedule their inspections according to what our measure finds.”
“The second thing we find is evidence for moral hazard. After a restaurant gets an A rating…often…they let the foot off the gas.”
Financial Networks, Predicting Potential Risk
Another thread of Mankad’s research is aimed at developing new statistical methods to analyze financial data. He models financial networks in order to identify communities, influential agents, and to characterize evolutions of the network structure over time. The goal is to measure systemic risk—the risk that an entire market or system could collapse—and be able to make informed decisions to avoid such an outcome.
Though Mankad’s academic background is not in finance or economics, he started working in the field during graduate school as a contractor for the Commodities Trading Commission. “My research brings a really different approach based on data integration and modern matrix factorization methods,” he says. “A lot of people have studied systemic risk. What we’re doing is different in nature, and the reason why is that my background is really coming from applied statistics. We're taking a data-driven and integration approach to trying to measure these things.”
In an NSF-funded project, Mankad and his collaborators—Celso Brunetti, Federal Reserve Board, and Jeffrey Harris, American University—are developing an integrative framework to identify and predict market participants that could endanger the financial system, for example, the housing market in the 2008 financial crisis. The model combines many data streams, including stock returns and interbank lending data typically only available to exchanges, as well as metadata and market announcements.
Mankad says that he hopes a tool like this would be useful to regulators like the Federal Reserve Board, the European Central Bank, and other large institutions responsible for ensuring that our financial worlds remain healthy. In the future, Mankad hopes to research more risk measures and develop tools that the Federal Reserve Board could incorporate onto their website, using his work to calculate risk on a day-to-day basis.
The Fun of Asking New, Impactful Questions
In all of his research Mankad’s goal is to create, discover, and solve something that’s useful to an organization and firm. He says that being on faculty at the SC Johnson College of Business has been especially beneficial to his work since he joined in 2015.
“There are awesome students at Cornell,” says Mankad. “My colleagues here are quite amazing. We have editors at all the major journals, and I can get great feedback from mentors. The school does a really good job helping junior faculty out.”
When it comes to doing research, Mankad says he is motivated by the fun of asking new questions. For example, with the hygiene project, he says, “To me, it's really cool there's no company really dedicating resources to looking into that issue. That's the role of academic institutions. We can look at problems over a long time or look at riskier projects. Then you find something that works and that can have a big impact.”