When three UC Berkeley students entered their computer science project last semester into a competition run by the business review site Yelp, they doubted they would win $5,000 and a meeting with the company’s representatives in San Francisco.
But on Sept. 27, James Huang, Stephanie Rogers and Eunkwang Joo received notice that Yelp had named their team one of four grand prize winners of the Yelp Dataset Challenge.
The trio sought to construct a project they could both enter in the Yelp competition and use for Computer Science 294, a class on behavioral data mining. After considering many topics, they chose to analyze Yelp restaurant reviews because “we like to eat,” Joo said.
The team started in early March and for the next two months used an online computer algorithm to analyze nearly 158,000 Yelp reviews of about 5,000 restaurants in Phoenix, Ariz., to identify the most important aspects of a dining experience. According to a UC Berkeley School of Information press release, the analysis pinpointed 50 different topics reviewers care about, including service and decor, which restaurants may in turn use to make specific improvements.
“This result is meaningful for Yelp and Yelp merchants as now they can dig into what customers care about and (know) specifically why customers are giving them high or low scoring reviews,” said Huang, who has now graduated from UC Berkeley and is working for a startup company.
Joo, now a graduate student at the School of Information, said that the team members didn’t expect to win the competition and was “surprised and thrilled” when they received the email notifying them that they had won.
According to their analysis, Asian restaurants tend to have the most polarized reviews of their service, with either strongly positive or negative comments. Customers also gave lower ratings to restaurants when they visited during restaurants’ busiest hours. The data shows a positive correlation between food quality and service, suggesting that reviewers’ perception of one may be influenced by their experience with the other.
“Our team was impressed by their methods and how useful this research could be for Yelp,” said Rachel Walker, a Yelp public relations specialist.
According to Rogers, a graduate student studying computer science, the results utilized an algorithm, called Latent Dirichlet Allocation, which automatically detects trends in thousands of reviews and calculates the results.
“This project is a great example of the work our students do: collaborating with students from other departments, requiring rigorous technical skills, and grounded in the real-world information needs of people and businesses,” said AnnaLee Saxenian, dean of the School of Information, in an email.
The team will be traveling to Yelp headquarters within the next couple of weeks to meet with Yelp representatives for a tour, lunch and photo shoot of the team receiving its $5,000 check.
Contact Michelaina Johnson at [email protected].