Project Name
Restaurant Success and Survival [Visualizations of Yelp Academic Dataset]
My Role
UX Researcher
Project Manager
Tools
- Tableau, GitHub, Excel
col 4 placeholder
Background
Due to COVID-19, many restaurants in the United States were impacted by government closures by different state regulations, employee layoffs, and loss of income. Our team had 10 weeks to develop a solution.
DESIGN QUESTION
How can we use Yelp’s academic dataset to help to better understand the different attributes that ensure restaurant success and or survival?
SOLUTION
Create an interactive website with data visualization dashboards using Tableau that will help restaurant owners and customers explore the elements that affect the closure of restaurants to make better decisions and understand the future trends of the restaurant industry.
The final project is viewable on restaurant-survival.vercel.app/
Discovery
We reviewed data from Yelp, OpenTable, and news articles relating to the COVID-19 and restaurant closures within YELP.
We found businesses providing home, local and professional services have been able to withstand the effects of the pandemic particularly well. Despite the bright spots in some sectors, restaurants and retail continue to struggle, and total closures nationwide have started to increase. In the wake of COVID-19 cases increasing, we see both permanent and temporary closures rise across the nation.
Research Questions
Our group focused on five different problem spaces to better understand the various factors that kept a restaurant successful
- 1) Are certain geographical areas more suited for restaurants than other areas?
- 2) Do review count and star rating influence restaurant success and survival?
- 3) Does review engagement help a restaurant's reviews and ratings?
- 4) Does the restaurant food category affect restaurant survival and success?
- 5) Are restaurants at specific price points more likely to survive?
Design Process
I began the design process by developing a deep understanding of our target audience. Through user research and a user-centered design approach, I identified key factors influencing user needs and explored multiple visualization concepts to determine the most effective chart types for our audience. This process involved numerous iterations, design evaluations, and refinements before arriving at the final visualization.
To inform the design, I conducted four informal user interviews with restaurant owners and customers to understand their attitudes toward restaurants and explore how those perspectives may have shifted during the COVID-19 pandemic.
After reviewing existing Yelp visualizations and prior research, I recommended a different strategic direction. Rather than focusing on time-series data, I chose to present a snapshot in time and prioritized interactive visualizations over static bar charts. This approach provided users with a more holistic understanding of the factors contributing to restaurant closures and enabled deeper exploration of the data.
PERSONAS
Based on user research we generated two personas (Restaurant Owner, Jill and Avid takeout Deliver User, Savannah).
The personas were used throughout the development process to help focus on our user needs and desires.
SKETCHES
For the initial sketches and prototypes my team and I explored a few graphs including symbol map, Tree-maps, Stacked Bar Charts and Gantt chart.
Dataset
I worked with two Yelp datasets: the Yelp Business Dataset and the Yelp Review Dataset. The business dataset contained 160,585 businesses represented as nested JSON objects with multiple levels of attributes, while the review dataset included more than 8.6 million user reviews.
Using Python, I combined the datasets, standardized and cleaned the JSON structure, removed irrelevant attributes, aggregated key variables, and flattened the data into a single-level format. I then transformed the dataset into a CSV file suitable for analysis and visualization. Through this process, I reduced the dataset to a curated set of 54,203 records, creating a clean and structured foundation for downstream analysis and interactive visualization development.
Mid-fidelity Prototypes
Building on insights from my research and close collaboration with designers and product stakeholders, I synthesized key findings into a series of visualization concepts and developed nine interactive prototypes. Through iterative design reviews and evaluation sessions, I explored multiple approaches to communicating restaurant performance data, ultimately focusing on visualizations that highlighted the percentage of open restaurants. This approach enabled users to better understand the factors associated with restaurant survival and long-term success while supporting more informed decision-making.
We used a choropleth graph (left) and a lollipop chart (right). This dashboard provides an approximation of how many restaurants are open in one area compared to other areas and offers an in-depth view using the ratio of open restaurants in each city.
Figure 2
Evaluation
To evaluate that our prototypes are working to achieve our design goals, I executed a usability study using our mid-fidelity and a few high fidelity interactive prototypes.
I recruited five participants which were a mix of restaurant owners and individuals who ate at restaurant(s) during COVID-19. I then incorporated a System Usability Scale (SUS) questionnaire that was administered using after each visualization interaction. At the end of each session, participants were asked to complet a semi- structured interview to debrief their experiences.
Final Design
After reviewing our participant’s feedback my team implemented iterations based on what we learned. We iterated on all of the graphs and created a more accessible chart by increasing contrast, removing filters, updating labels, and changing some graph types. The goal of our overall visualization is to help restaurant owners and customers determine the factors that help their restaurants succeed. We chose to create coordinated multiple views to maximize storytelling and link multiple factors in one visualization. Applying Ben Schneiderman’s taxonomy for visualization tasks, we started our multivariate visualization by providing an overview of the major geographical areas covered by Yelp’s dataset, breaking down different variables that lead to restaurant success, and finally tying together all the variables (i.e. state, city, cuisine etc.)
All visualizations provide details-on-demand through a tooltip with exact details and incorporate brushing/linking techniques to allow users to highlight a single data point to better dive into relevant information. All of our visualizations use the history technique, in order to check and uncheck filters to return to the original state of the visualization. Since all visualizations are stored on Tableau Public, this gives our users the ability to extract the for their own business decisions and personal use.
Challenges
If I had more time, I would like to focus on three areas: collect time-series data, more user research with different various cuisine types , and exploration of other visualization tools. Yelp does not offer time-series data, which limits our scope to a snapshot in time. If I were able to extend the length of this project, I would save data from Yelp periodically. With more time-series data, we would better understand the impact of COVID-19 on restaurants and identify more trends for the survivability of restaurants.
Although Tableau is a powerful tool, the visualizations were slow for testing and often had issues loading. In the future, I would like to try D3, Python, Pandas, and R to build more complex interactive visualizations.
Lessons Learned
Effective communication and clear roles are key to teamwork success. At the early stages of the project I found my team members having very different ideas and directions on how we should design our project. I helped in creating Ideation sessions to allow my stakeholders to share their individual ideas. From there I held voting sessions to confirm the research questions that we were most passionate about.
Overall I had the pleasure of working with a very dedicated and collaborative team.