UCLA Anderson MBA students conduct Applied Management Research field study projects in lieu of a thesis. Anderson gives full-time students the Business Creation Option in which entrepreneurs launch companies while still in school. Team Connoisseur researched the restaurant landscape in L.A. to devise a delicious formula for picking local dining options that improves on standard rating sites.
By Zodin Del Rosario (’17), Richard Fernandez (’17), Kimberly Kalb (’17), Christina Rath (’17) and Sean Ginley (’17)
The restaurant search process is broken. With so many restaurants in L.A., people feel overwhelmed, and current websites like Yelp prove untrustworthy and cumbersome for people to sort through.
Connoisseur is a mobile app designed to offer personalized restaurant recommendations based on your dining history.
To gather seed data for our algorithm, the Connoisseur team surveyed 82 respondents to rate all the restaurants they had visited in Santa Monica. We explored multiple methodologies for predicting restaurant scores, including a k-nearest neighbors algorithm (k-NN), in which users are plotted in vector space and the actual physical distance between users is measured. In a k-NN, users with similar restaurant ratings will be closer to each other. We looked to find the five nearest neighbors to any user. Taking the average ratings of these five nearest neighbors, we found the other restaurants that these similar users loved.
We were able to test the viability of this approach by removing some of the data and using the algorithm to predict the deleted scores. This method was most effective when users had rated at least 25 restaurants, meaning once we moved outside of Santa Monica we would need users to rate even more in order to give accurate predictions.
We attempted to reduce the barrier to entry by clustering users into groups rather than comparing them to all other users. Using a technique called k-means clustering, we grouped users based on their ratings of just five restaurants that were frequently visited but had varying ratings. We had 95 respondents rate the five restaurants, and from there grouped them into four clusters. Each cluster received a separate email prompt with 10 restaurants we believed they would like, and we asked them to respond and let us know if they had visited any of these restaurants and liked them.
Using this test, we were able to recommend restaurants that scored a 4 or 5 more than 80 percent of the time. The results indicated that this approach could successfully provide personalized recommendations for users. We could build the algorithm to be more accurate by incorporating details such as cuisine type, neighborhood and cost. We anticipate there will be some barriers to expanding the clusters across the city and may need to add more groupings.
Instead of investing resources immediately into making an app, we first created a mobile-optimized website, which is now available on diningconnoisseur.com. Here we are able to collect contact information for beta testers as well as restaurant ratings throughout Los Angeles, which will further help with seeding and testing our algorithm.
We have added fellow FEMBA Ray Xiao to our group to spearhead our mobile app development and work with Sean Ginley to implement and refine our algorithm. At the same time, we have developed a social media presence and continue to conduct outreach to potential customers and partners in the restaurant industry.
Follow Team Connoisseur!