Nara Logics was founded in 2010 to mine the deep web and uncover relevant information for businesses to offer their customers. The company, founded by MIT neuroscientists, recently landed $7 million in funding, and makes money through revenue sharing with partners such as Uber, OpenTable, and TripAdvisor, and licensing its AI backend to vendors like Singapore mobile provider SingTel.
In the process of indexing venues for American cities, the team made an unexpected discovery: There are a few “supernode” restaurants that connect every eating establishment in large cities. Specifically, the top 1% of restaurants in large cities like New York, San Francisco, and Los Angeles are connected to between 60-70% of all the restaurants in their city. Although they aren’t necessarily the top-rated restaurants in a market, they do tend to share the most connections to other restaurants both among employees and customers, as well as other external sources.
Nara’s software builds a connection graph for individual venues that links them together based on their properties. The company’s web crawler vacuums up structured data, like restaurant listings, and unstructured data (reviews, blog posts, newspaper articles) to build profiles of the business establishments. These points are then linked together based on their characteristics. Scores for individual restaurants are generated for each individual user, based on the ratings Nara’s users gave for other restaurants.
It’s a massive index that can be queried along almost any dimension. “For restaurants, you could ask for the current best sushi restaurant in the world or for the best restaurant with a waterfront view close to the Financial District,” says cofounder and CTO Dr. Nathan Wilson. “It’s a brain-like system that sits on top of any large data set, and automatically organizes this information for personalization and recommendations.”
By running their own analysis, Nara learned that these supernode restaurants aren’t only linked by customers mentioning them in the same places, they share a much-higher-than average number of social connections to other restaurants as well. Interestingly, these restaurants were connected to both expensive and inexpensive restaurants through social connections. At least 15% of the restaurants they’re connected to are “low-priced options,” Wilson says.
These machine learning and neural network techniques aren’t specific to just lodging or restaurants. Nara’s approach bears a strong resemblance to the system Pandora uses to find connections and similarities between different songs, for instance. In one example Wilson gave, the company’s web recommends restaurants to users based on similar decor or physical setting (such as a waterfront view) rather than strictly using cuisine, price point, or neighborhood as a primary metrics. Nara’s hope is that machine learning and data science can find connections between restaurants that human reviewers can’t–and that it can build a better recommendation in the process.