Using Data Analysis and Natural Language Processing to Optimize Success in New Restaurants
TLDR
Restauranteering is a tough business to get into since it has higher than average failure rates. Here we identify characteristics of highly rated restaurants to optimize chances of success in new restaurants. EDA reveals that population demographics of the location of the restaurant are vital for the success of a certain cuisine type (i.e. education positively correlated with Japanese cuisine). Topic modeling shows that people valued atmosphere more than decor, and most appreciated fresh ingredients for meat, fish, and alcoholic beverages. Customer service and taste were the two most central topics in a 5-star restaurant. Freshness and Vibe/atmosphere were not too related, meaning if a restaurant provides high quality ingredients, they can get away with not having the perfect atmosphere, and if a restaurant has a fun and lively atmosphere, they did not need to rely as heavily on good quality ingredients to score a 5 star review.
Figure 0: (Top) Distance map of topics and their relativity to each other. (Bottom) Tokens that make up each topic.
Overview
Restaurants are one of the toughest businesses to go into, with 26.16% of restaurants failing within the first year, anda cumulative failure rate around 60% by year 3 ¹. It has been shown that yelp rating has an effect on revenue, with a 1 star increase leading to a nearly 10% increase in revenue ². We can use the yelp dataset to identify characteristics of highly rated restaurants to optimize chances of success when opening new restaurants. Here, we’ll look at 5 cities in Arizona with the highest number of 5-star restaurants.
EDA
When exploring the data, we can ask a few intuitive questions: What cuisines do well in each city, and how do the demographics of each city affect the success of that cuisine? Figure 1 gives us a baseline of demographics of 5 cities in Arizona. The cuisines with the highest number of 5-star restaurants in these 5 cities are Breakfast & Brunch, Japanese, and Mexican. We can explore each of these cuisine categories individually to pick out trends.
Figure 1: Demographics of Phoenix, Scottsdale, Mesa, Tempe and Chandler Arizona. Mesa and Chandler have less 5-star restaurants per capita than the other cities, with statistical significance.
Studies have shown that “Brunch” has become wildly popular amongst demographics of young, middle class, white citizens⁵ ⁶. If we look at our demographics baseline in Figure 1, we could hypothesize that Tempe would be the city most likely to have the highest percentage of successful brunch restaurants. Figure 2 confirms this.
Figure 2: Percentage of brunch restaurants in Tempe, Chandler, Mesa, Scottsdale and Phoenix.
We can make similar predictions using literature reviews for Japanese restaurants. There is a positive correlation between education levels and Japanese restaurant density. In fact, the best indicator for the prevalence of Japanese restaurants in America is education density. Ethnic composition and income did not have a significant effect on restaurant density ⁷. When we look at figure 3, however, we can see that Tempe again has the highest percentage of Japanese restaurants- even though Scottsdale has the highest density of educated residents. Upon inspecting the demographics, we can see a clear confounding variable: Scottsdale has much older residents on average. Studies have shown that older Americans are much less likely to consume sushi ⁸, so when we adjust for this it actually makes sense that Tempe would have the highest density of Japanese restaurants.
Figure 3: Japanese restaurant density in Tempe, Chandler, Mesa, Scottsdale and Phoenix.
Education levels have shown to have a negative correlation with density of Mexican restaurants⁷. This is confirmed in Figure 4, with statistical significance. Phoenix and Mesa have the lowest education density, and the highest Mexican restaurant density.
Figure 4: Mexican restaurant density Tempe, Chandler, Mesa, Scottsdale and Phoenix.
It’s clear that it’s important to take location demographics into account when looking into opening a new restaurant.
Natural Language Processing
The Yelp data set includes restaurant reviews/comments from users. We can clean and pre-process the review data (removing punctuation, capitalization) and turn the data into tokens. Stemming and lemmatization could be considered, but here it is suggested that might do more harm than good. We can also remove stop words, which won’t be useful to the analysis. Now that we have our cleaned tokens, we can add bigrams to give the tokens more meaningful interpretation (i.e. “goodfood” vs “good”).
Once the data is cleaned and preprocessed, we can continue to topic modeling shown in Figure 5. The closer two topics are on the distance map, the more related they are to each other. We can draw some meaningful conclusions from this:
It seems like when it comes to freshness, the most important ingredients to have fresh are meats, fish, and alcoholic beverages.
Service with good managers and good serves was the most vital.
People prioritized vibe/atmosphere over restaurant decor.
It was important to have a high range of flavors (spicy, sweet, sauce).
Service and taste were pretty central topics in high reviews, but what's interesting is that freshness and vibe were not. Vibe is most strongly related to Service and staff. This could be interpreted that if restaurants have great staff, great taste in their food, and a great atmosphere they might not need to have super high quality ingredients (i.e. a dive bar with live music and great wings). The same is true for the freshness topic, if a restaurant has very fresh and high quality ingredients, they might need to rely as much on having a fun vibe and great atmosphere.
Figure 5: (Top) Distance map of topics and their relativity to each other. (Bottom) Tokens that make up each topic.
Insights/Suggested Actions
Demographics(age, race, education) play a heavy role in rating of a restaurant. Therefore, it’s important to select a location with a high density of people in your target demographic that would be attracted to the type of cuisine you are selling.
Focus budget on building a great customer service team and excellent taste in food. Then, select to prioritize either quality of ingredients or curating a fun atmosphere. This may depend on your cuisine- sushi restaurants should prioritize freshness, while brunch restaurants may want to prioritize atmosphere. If your restaurant offers meat, fish, and drinks be sure that these are the ingredients you get fresh.
References
https://journals.sagepub.com/doi/pdf/10.1177/0010880405275598
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=1928601
https://bestneighborhood.org/race-in-phoenix-az/
www.census.gov
https://books.google.com/books?hl=en&lr=&id=igYHBAAAQBAJ&oi=fnd&pg=PA93&dq=brunch+and+college+towns&ots=cXazrTwBSM&sig=bhiTyWZobBte7ZsA9OH4RXXmo6g#v=onepage&q&f=false
https://journals.sagepub.com/doi/full/10.1177/0001699313498263
https://www.tandfonline.com/doi/pdf/10.1080/15528014.2017.1337390?needAccess=true
https://www.keltonglobal.com/recognition/pei-wei-releases-sushi-survey-results-adds-new-rolls/
https://journals.sagepub.com/doi/pdf/10.1177/0010880405275598
https://journals.sagepub.com/doi/pdf/10.1177/0010880405275598
https://medium.com/broadhorizon-cmotions/nlp-with-r-part-3-using-topic-model-results-to-predict-michelin-stars-ba8ec1b182c2
Github
https://github.com/LaurenTomlinson/R_Project
About Author
