So you want to open a restaurant?

An analysis of a Yelp Academic data set with a focus on restaurants.

Yelp has changed the way people decide which restaurant to go to. The star ratings can make or break a business.

The Yelp academic data set contains 5,996,996 reviews, 188,593 businesses, 280,992 pictures, and 10 metropolitan areas.

The business data includes the names, longitude and latitude, review count, average stars (rounded to the half star), whether the business is open or not, categories (that are not mutually exclusive), and attributes (does takeout, parking, etc).

These are the questions that will be examined:

Are there differences in the average stars by major Yelp categories (Restaurants, Shopping, Nightlife, …)?

Restaurants comprise approximately 30% of the businesses in the data. This is followed by Shopping (16%) and Night Life (7%).

The chart shows the average stars by the type of business. The red line indicates the overall average (3.63). The average stars for restaurants is 3.44. It looks like people are more critical of restaurants than other types of businesses.

Looking at the top 20 most frequent restaurant types, cafes seem to have the highest average stars; fast food has the lowest. The red line indicates the average for all restaurants.

To answer this question, we trained a machine learning model using most of the data and then checked our predictions on the remaining part. The model was able to predict if a business is open or not with an F-score of .91 (with 1.0 being perfect).

Awesome! But what factors help determine if a business will close or not? Looking at features that were most import to the model, we see that the top three features were longitude, review count, and latitude.

Does the density of restaurants affect the star rating? Have you ever noticed that restaurants tend to be located in clusters? Does this actually help them?

To calculate the density, the latitude and longitude for each restaurant was used to determine its S2 cell at level 13. S2 cells divide the globe into roughly equal regions. A level 13 S2 cell has an area of approximately 1.27 sq. km.

Then for each S2 cell, the count and average stars were computed. The densest cell had 284 restaurants and is in downtown Toronto. The average stars was 3.32.

The correlation between count and average stars was found to be -0.03. So it appears there is no correlation between density and star rating.

To sum it up:

However, without knowing how the data was sampled or if the data was curated in any way, we should be cautious about applying these observations to the general case.

Related posts:

Some like to use scare tactics to say that advertising costs are too expensive and there are too many businesses running Facebook ads these days. If you sign up for a 14 day Clickfunnels free trial…

I think we all know that vacuuming keeps our floors and our homes cleaner, so it’s a good thing to do…and even more important if you have little ones in the house. But very few people realize the…

Mindset is not just about effort; the old adage that if you put your mind to something that you will achieve it. Effort is key, but there is more to it. In the Journal of Clinical Oncology, they…