ADVERTISEMENT

The size and racial makeup of a city, the price of a meal and even the weather can skew the quality and quantity of online restaurant reviews, according to the first large-scale academic study to analyze how outside factors affect crowd-sourced review sites.

The recently released study used computer models to examine nearly 1.1 million reviews of 840,000 restaurants over nearly a decade.

Online reviewers posted most often in July and August, researchers said, but those reviews were more likely to be negative. And regardless of season, if the weather was uncomfortable, all the worse for the chef. The most negative reviews were written when it was colder than 40 degrees or warmer than 100 degrees, or if it was raining or snowing.

The researchers, Saeideh Bakhshi, a doctoral student at the Georgia Institute of Technology; her husband, Partha Kanuparthy, who works for Yahoo Labs; and Eric Gilbert, an assistant professor at the university, said the weather’s seeming sway over reviews surprised them the most.

They also found that some regions of the country were more prolific in their online reviewing habits than others. Restaurants in the Northeast and on the West Coast were reviewed more than those in the South or the Midwest.

Predictably, urban areas with a higher level of education and income tended to participate in online review sites more often.

Among the other findings: People who waited a long time for a table in busy cities were more forgiving than those who waited in smaller communities. And sushi restaurants were consistently rated higher than hamburger places, the researchers said, showing that ambience and a higher meal cost could produce better reviews.

“That speaks to the perception of price,” Bakhshi said. “Places that have nice ambience and are listed as romantic or trendy or more expensive, the rating is higher.”

The couple came up with the idea for the study when they were in Seattle, where Kanuparthy was working for Microsoft. They noticed that Seattle reviewers seemed to give fewer stars to similar restaurants than reviewers in Atlanta.

Certainly, the respective palates of each city could be a factor, but the couple thought there might be a way to measure other differences. So they created analytic computer models based on data from several sites, including TripAdvisor, Foursquare and Citysearch.

Yelp, which has posted more than 53 million reviews since it began in 2004, would not release enough of its data, they said, and thus was not included.

The team found that Seattle, as a whole, tended to offer lower reviews than many other cities. Sunny San Diego had the most five-star reviews.

Overall, the researchers said, regional differences must be taken into account when considering the validity of a review. A thousand positive reviews of a Chinese restaurant in San Francisco may mean more than a thousand in Atlanta, which is not known for its Chinese cuisine.

“We have to consider the domain of knowledge,” said Bakhshi.