Analysis of the Uber Eats Recommendation System

Author

Tenzin Thakuri

Introduction

Uber Eats is an online food and grocery ordering and delivery platform operated by Uber Technologies. It allows customers to order meals, groceries, drinks, and other products from local restaurants and stores through its website or mobile app. According to Uber’s official Help Center, Uber Eats is available in more than 6,000 cities across 45 countries worldwide. It makes sense for me to perform Scenario Design twice,once from the organization’s perspective and once from the customers’ perspective because Uber Eats has different goals for its business and its customers.

Scenario Design from the organization’s perspective (Uber Eats)

Who are the organization’s target users?

Uber Eats operates a three-sided marketplace:

  • Customers: People who use the platform to discover and order food.
  • Restaurant partners: Restaurants that use Uber Eats to reach customers and receive orders.
  • Delivery partners: Couriers who pick up food from restaurants and deliver it to customers.

What are their key goals?

From the organization’s perspective, Uber Eats wants to:

  • Increase the number of customer orders.

  • Improve customer satisfaction and retention.

  • Increase revenue

  • Give restaurant partners opportunities to reach more customers.

  • Maintain an efficient marketplace by balancing restaurant demand and delivery availability.

How can the recommendation system help?

The recommendation system can help Uber Eats achieve these goals by:

  • Recommending restaurants that match customer preferences.

  • Giving new and less popular restaurants opportunities to appear in recommendations.

  • Ranking restaurants based on customer interest

  • Considering delivery availability and estimated delivery times.

  • Helping customers discover a wider variety of restaurants.

Scenario Design from the customers’ perspective

Who are the users?

The users are customers who use the Uber Eats website or mobile app to order meals, groceries, drinks, and other products from local restaurants and stores.

What are their goals?

  • Find restaurants that match their food preferences.

  • Discover new restaurants and dishes.

  • Find affordable food within their budget.

  • Compare restaurant ratings and delivery times.

  • Save time when searching for food.

  • Receive their orders conveniently and on time.

How can the recommendation system help?

  • Suggesting restaurants based on their previous orders and food preferences.

  • Recommending dishes based on their interests.

  • Showing restaurants available in their delivery area.

  • Considering the time of day and location when making recommendations.

Reverse Engineering Uber Eats recommender system.

How Uber Eats displays recommendations.

The Uber Eats homepage displays different recommendation sections to help customers discover restaurants and food options.

  • Featured on Uber Eats: Displays selected restaurants in cards on the homepage.

  • Order Again: Shows restaurants that I have ordered from frequently, making it easier to order again.

  • Deals to Explore: Displays restaurants and food items with available deals and discounts.

  • Tailored to Your Taste: Shows personalized restaurant and food recommendations.

  • Popular in Your Area: Displays popular restaurants near my delivery location.

How the recommendation system may work behind the scenes.

1. Collect information

Uber Eats uses customer and restaurant information, including:

  • Previous orders and food preferences.

  • Customer location.

  • Time of day.

  • Restaurant availability and delivery time.

  • Restaurant menus and prices.

2. Two-Tower Embeddings

Uber uses a machine-learning architecture called Two-Tower Embeddings (TTE) to retrieve relevant restaurants, dishes and grocery items in milliseconds across millions of candidate options.

  • One tower represents the customer (eater).

  • The other tower represents the restaurant (store).

  • The system compares their numerical representations to identify relevant matches.

    Figure 1. Left image shows the architecture of eater-item two tower embedding model; Right images shows how we can use the two tower embeddings to find the most relevant restaurants for eaters.

3. Candidate generation

Uber has many restaurants on its platform. Instead of evaluating every restaurant in detail for every customer, the system first retrieves a smaller group of potentially relevant restaurants

For example, if a customer likes pizza, the system may retrieve nearby pizza restaurants that are available for delivery.

4. Personalized ranking

After selecting the potential restaurants, Uber Eats ranks them based on how relevant they are to each customer using a framework called Multi-Objective Hierarchical Recommender (MOHR)

It considers four objectives:

  • Customer conversion: Encouraging customers to place an order.

  • Customer retention: Encouraging customers to return to Uber Eats.

  • Basket value: Increasing the value of customer orders.

  • Restaurant exploration: Helping customers discover new restaurants.

5. Display and update recommendations

The final recommendations appear on the customer’s homepage.

Recommendtaions for Improvement

  • Nutritional-based recommendations: Include calorie, protein, fat, and carbohydrate information so customers can find meals that match their dietary goals.

  • Dietary preference filters: Allow customers to personalize recommendations based on vegetarian, vegan, halal, gluten-free, and other dietary preferences.

  • Budget-based recommendations: Allow customers to set a preferred budget and recommend meals within that price range, including delivery fees.

Conclusion

Earlier, Uber Eats used Deep Matrix Factorization (DeepMF) to retrieve restaurants for its homepage. However, it had limitations, especially in scalability and maintenance. Uber introduced Two-Tower Embeddings (TTE) to provide a more scalable solution that can find restaurants and food items matching customer interests while supporting additional features. In addition, the Multi-Objective Hierarchical Recommender (MOHR) helps organize and rank restaurant recommendations while balancing different business objectives.

Although Uber Eats already provides personalized recommendations, it could improve the customer experience by adding nutritional-based recommendations, dietary preference filters, and budget-based recommendations. These improvements could help customers find food that better matches their needs and preferences.

References

Uber. (2026, April 14). Innovative recommendation applications using two-tower embeddings. Uber Engineering. https://www.uber.com/us/en/blog/innovative-recommendation-applications-using-two-tower-embeddings/

Uber. (n.d.). Uber Eats: Recommending for a marketplace. Uber Engineering. https://www.uber.com/us/en/blog/uber-eats-recommending-marketplace/

Wang, Y., Tao, L., & Zhang, X. X. (2024). Recommending for a multi-sided marketplace: A multi-objective hierarchical approach. Marketing Science. https://doi.org/10.1287/mksc.2022.0238