Helping Restaurant Owners

Mining Information From Yelp! To Improve Restaurant Success

According to a recent CNBC article, 60% of all restaurants close before their first anniversary, and 80% before their fifth. There are over 624,300 restaurants in the U.S., with total annual food and drink sales estimated at $782 billion, which supports a workforce of approximately 14 million individuals. (

We set out with a single objective in mind, utilizing Yelp's rich data set to create a tool that can help restaurant owners improve the likelihood of their success.

After a number of lengthy interviews with restaurant owners in different cities, we began to narrow our focus on several questions that we believed could be answered by using the Yelp dataset, as well as some additional neighborhood-level demographic data sets.

  • How are my individual dishes being received by my customers?
  • What restaurants are most similar to mine?
  • How can I better analyze potential restaurant locations?
  • How are reviews trending for similar restaurants?

While developing answers to these questions, we found a few other useful (and interesting) things along the way...

YaBO! - Pittsburgh, PA

MVP Release 1.0 - "City of Bridges"

More Information on YABO!'s MVP Release

The MVP Release of YaBO! includes all restaurants in Pittsburgh, PA, U.S., that have been reviewed on Yelp! over the past 12 years. The YaBO! prototype utilizes data from Yelp! and the US Census Bureau. The data used for the Pittsburgh MVP release includes (approximately):

  • 2,733 Restaurants
  • 115,789 Reviews
  • 2,480 Menu Items

YaBO! utilizes the following techniques and technologies to provide the information within this web portal:

  • Paralax Web Portal with Javascript and JQuery
  • Regression Analysis
  • Natural Language Processing
  • Machine Learning
  • Geolocation Cluster Mapping

YaBO! - 2017 "City of Bridges" Release

YaBO! Business Insights

The 4 Dimensions of YaBO!


Understanding Your Market

In order to differentiate your business, you must uderstand who your competitors are, where they are, what they are serving, and how reviewers are receiving their offerings.


Understand Your Customers

Your menu must be developed with an eye toward your local neighborhood demographics. Also, you need to understand the difference between the reviews you receive from casual Yelp! users vs. Elite Yelp! users.

Micro-Industry Trends

Understanding Local Trends

The Restaurant Industry is extremely dynamic. Understanding how changes in the in the frequency and quality of your reviews compares to those in your local industry can provide important insights into the future.

Cuisine Clusters

Understand Cuisine Competitive Locations

It is normal to see cuisine-specific "Hot Spots" develop in a market, but you must be aware of these when looking to open your first restaurant, or to expand to an additional location.

Understanding Yelp! Reviewers

The Difference Between Casual and Fanatical Yelp! Users

The Love/Hate Relationship with Yelp Newbies

As a Restaurant Owner, it is helpful to understand as much about Yelp Reviewers as possible so that you can put their feedback into as much context as possible. The graph below illustrates the relationship between how active a Yelp Reviewer is, and how they rate restaurants. In the graph, Yelp Reviewers are divided into 10 categories based on the total number of reviews they have posted in the past 12 years. Take a minute and explore the Review Ratings that these different categories of users tend to post.

1.) Start by clicking on the bar for the single (1) Review users on the far left. Look at the horizontal bars on the bottom and see how their reviews are distributed. Notice that there is a distinct Bi-modal distribution. Intuitively, this makes sense when you think that these people have likely never used Yelp! before, and something has prompted them to post a very good, or very bad review.

2.) Next, click on each group with an increasing number of reviews and notice how the distribution changes as you move toward the Elite Yelp! users who are posting more reviews. Notice that as you get to Reviewers who have posted at least 4 Reviews, the distribution of their Ratings is skew left with 5-star reviews as the most freqent, and lesser star reviews occuring less frequently in sequence.

3.) And as you move even further towards users who have posted 10-24 Reviews, you'll notice that they are now splitting most of their reviews between a Good and a Very Good Rating.

4.) But as soon as you look at users who have posted 25 or more reviews, you'll instantly notice a shift in their ratings. These 'Super' users (many of them earning the 'Elite' title from Yelp), are more likely to point out both positive an negative attributes about a restaurant. And in giving a more balanced review, many of their reviews tend to be 4-star. And as you move up to the 100+ Review Users, you see that 3 and 4-star reviews are the most frequently used in their reviews. In contrast, go back and click on the single (1) Review Users and notice how different their reviews are from the 100+ Review Users.

"Wife Hated The Oily Pizza And Rude Service!"

Word Frequency Across The Best & Worst Reviews

Visualizing Word Frequency in Good and Bad Reviews

These visualizations (a.k.a. "wordclouds") were generated by identifying words that most predicted whether a review was good or bad. While wordclouds aren't exactly 'scientific', they can provide interesting insights into common positive or negative review comments within a cuisine category, which may help add context to reviews that your restaurant has received. Everytime we look at these word clouds we discover something new, but here are a few interesting finds to get your search started:

  • Seafood -> Fry, Price, BYOB, Scallops, Brunch, Decor
  • Burger -> Wife?, Truffle, Free, Onion, Homemade, Friendly
  • Bars -> Craft, Attentive, Money
  • Italian -> Rude, Calamari, BYOB, Homeade
  • Pizza -> Oil, Rude, Cold, BYOB
  • Mexican -> Chips, Grab, Fry, Outside

Method: Utilizing a count vectorizer and WordNetLemmatizer in Python, we identified the top restaurant categories for Pittsburgh, PA, as well as the 40 words in the one and five star reviews for each of these top categories. These values were ranked and loaded into Wordle to generate wordclouds using a color palette with reds and blues to indicate the poor vs. good review words and size to indicate relative frequency.

Select a cuisine:

Word Cloud by Cuisine

"Whose Reviews Are Improving?"

Explore Pittsburgh's Reviews By Category

Looking at Reviews Over Time By Restaurant Category

The following graph provides a unique vantagepoint into how reviews are trending for restaurants within (and across) cuisine categories. For instance, you can select your cuisine type and quickly see which restaurants have had and increase (or decrease) in both the frequency and quality of their reviews over the past year.

Instructions For Using This Dashboard:

1.) Begin by selecting the cuisine category you would like to explore. You will first need to clear all of the categories by selecting (All) at the top of the list on the right-hand side of the dashboard.

2.) Once you have chosen your category (or multiple categories), you can begin exploring the scatterplot. Here are tips on how to interpret it:

  • Colors - the color of the points on the graph indicate the restaurant's total average review. Blue is 4 or 5, Grey is 3, and Red is 1 or 2.
  • Size - the size of the point indicates how many total reviews this restaurant has ever received.
  • X-axis (moving from left to right) - This indicates the amount of change in the number of reviews the restaurant has received this year compared to last year.
  • Y-axis (moving from top to bottom) - This indicates the amount of change in the average of reviews this year compared to last year.

Typically, restaurants in the top right quadrant have experienced an increase in both the number of reviews and the quality of reviews this year over last year. Restaurants in the lower left quadrant have experienced a decrease in the number and quality of reviews over the past year.

3.) Once you identify a point of interest, you can click on it to take several possible actions:

  • View summary statistics in the pop-up menu
  • View the Quarterly Review Averages for both the Quality Ratings and the Number of Reviews for the past several years
  • View where the restaurant is in the city, using the map in the lower right-hand side of the dashboard
  • Click on 'hyperlink' in the pop-up menu to go to the restaurant's Yelp! page to do further investication

Now go see how your competitors are trending over time!

Cuisine Cluster Analysis

Exploring Pittsburgh's Restaurant Distribution

Visualizing Consumer Insights And Strategic Restaurant Placement

This map tool pulls in restaurant and demographic data and creates overlays for various features such as mean neighborhood income and restaurant density. The graph provides summary statistics on a granular level, so the user can get insights on both the residents and the food scene for any given neighborhood. For a prospective restaurant owner, the map enables filtering on price, cuisine, and restaurant name, so they can easily identify the best location for a new restaurant and what clientele to design the menu around. In addition, each restaurant marker links to a custom insights dashboard for that specific restaurant.

How can you use it?

  1. Understand the neighborhood in which your business resides.
    • Expensive restaurant in a low-income neighborhood? Offering lower-cost menu items could attract new business.
    • Is the average neighborhood age 55 years old? This may not be the best spot for a trendy new acai cafe.
  2. Evaluate the competition.
    • Busy area with no Chinese restaurants? The location may be perfect for meeting the needs of an untapped market segment.
    • A neighborhood overstuffed with 4-star restaurants could be a great place to provide an affordable sandwich shop.
View The Most Recent Listing of Restaurants With Available Menu Analysis

Yabo! Architecture & Quality Measurements

For More Information, Click "How It Works"