The best burger - an analysis based on reviews of delivery restaurants

It is striking that people often have strong opinions about where to eat the best burger. Or the best Roti. You then get a colorful story about the one place that serves towering heart-attack burgers, or just purple vegan burgers overflowing with fresh ingredients. Usually accompanied by a good association: "This is where I always go after going out" or "the owner is an incredibly nice guy." Of course, it is a very personal and limited consideration of what is the best burger. That's why we posed the question with some of our colleagues at Cmotions: is there a data-driven method to determine where to eat the best burger?

To answer that question, we need (1) a source of reviews, (2) we need to distill which reviews are about the desired product, and we need to (3) soundly aggregate them to a final verdict for the restaurant. With a prerequisite: a limited distance we are willing to travel for a feast in our mouths.

1. Source of ratings

Instead of asking people around us, we use the millions of reviews that customers post on the website of a well-known delivery service. Because of uncertain opening hours during the pandemic, it has become very interesting for restaurants to join a delivery service. This is reflected in the growth in the number of restaurants delivering through meal delivery service websites. That makes this a rich source of reviews, although not every restaurant is affiliated. Michelin star restaurants and other "haute cuisine" probably fall outside the equation (but they also rarely serve a burger the way we imagine a burger to be).

2. Assessment of the desired product

Customers can leave a review by giving a number of stars (1-5) and optionally explain their judgment in writing. The reviews give a nice idea of the quality of the food. However, we are curious about the quality of a specific product. Therefore, for each review we looked to see if the product in question was mentioned, and stored those reviews separately. The reasoning is that when someone mentions a product, we can state with a higher degree of certainty that the review was also about the burger, and not about another dish offered by the same restaurant. To illustrate, of the 7.2 million reviews, only 35% contain a commentary. The remaining 65% of reviews consist only of stars with no written review. If we search specifically for reviews that contain the word "burger," we are left with about 72,000 reviews, about 1% of the total so.

3. The formula for final judgment

With a weighted combination of overall and targeted (i.e., one named the burger in the review) ratings, we come a long way with an overall best burger verdict. After the first iteration, we saw that this still did not give a satisfactory picture. Is the burger from a restaurant with an average review rating of 4.6 stars based on 100 targeted reviews better than the burger from a restaurant with an average of 4.5 stars based on 700 targeted reviews? The proportion of reviews per restaurant that are specifically about burgers gives an indication that the burger is an important or popular dish for that restaurant. Combining these elements, we arrive at the following formula:

Where the w represents the weight. That is, how much a variable counts in the final score. The variables are scaled (after division) before they are weighted and summed. So basically you can say that the burger metric represents how many standard deviations the restaurant is above or below the mean. Where the average is then a weighted combination of the average number of stars, the percentage of targeted reviews and the number of targeted reviews. The weights were chosen as follows:

The ratings of reviews where a burger is specifically named weigh most heavily (80%) in the final rating. The percentage of reviews where the burger is named weighs 15% and the total number of targeted reviews weighs 5%. There are also prerequisites before a restaurant is included in the comparison. It must have at least 100 reviews, of which at least 30 reviews mention the burger. Applying these criteria, nearly 50,000 reviews at 611 different burger restaurants remain in our dataset. So please note, that new burger bar around the corner that everyone is talking about is not included in this model.

The result

We sorted each restaurant by its burger metric to compile a top 10 in the Netherlands. Hungry Jake from Hilversum scored the best! Figure 1 shows which other restaurants are worth an order and based on which data the burger metric was calculated.

Wondering in which city you order the best burgers? See below Figure 2 how the score of burger restaurants is spread by city. Rotterdam has the largest spread. In Enschede you are almost always above average.