top of page

Yelp Redesign: Artificial Intelligence

Integrating AI into Yelp to improve customer experience
Figure 12: Final Iteration of "Messaging Yara"
final iteration.png
Final Design of Yelp with our recommender algorithm
final iteration 2.png
Final Design of "Messaging Yara" (our recommender algorithm)

Overview

The Challenge 

​

​

Proposal 

​

​

​

​

Value Proposition

​

​

Why? 

​

​

​

​

Skills and Tools 

​

​

My Role 

Will adding an intelligent user interface to Yelp add value for stakeholders? And if so, is it feasible?
​
Adding an intelligent user interface (IUI) to Yelp’s mobile application is feasible and will add value for stakeholders. An IUI can leverage user’s past transactions and interactions, crowd-sourced data, and the large amount of unstructured data already in Yelp to provide personalized recommendations in a natural and friendly manner.
​
Instantly provides customers with a short and tailored list of restaurant options, with adequate reasoning, that keep their specific taste requirements in mind.
​
Adding 'intelligence' to Yelp's interface can:
  • Prevent satisficing and assist in optimal decision making by reducing the cognitive load placed on the users’ working memory.
  • Decrease the time spent searching for a restaurant that meets their needs.
​
User Interviews, Intelligent Architecture Flow Chart, Wireframing, High-fidelity Prototyping, UXPin
​
Worked as a UX Researcher and Designer on a team of 3. I conducted 2 user interviews, synthesized the data, and created the intelligent architecture flow chart. I helped with mapping the current user flow and creating the wireframes and high-fidelity prototypes.
​

Analysis of Yelp

problems with yelp.png
Figure 1: Summary of the problems experienced when using Yelp to locate a restaurant
Analysis of Yelp

Understanding the Users

User Interviews

User Insights 

  • Users look for and trust recommendations from friends or family
  • They expect personalized search results
  • Scrolling and reading through all the current search results on yelp was described as time-consuming and frustrating.​
​
User Interviews
Quote.png
Figure 2: Quote from Participant #3

Mapping the Current User Flow

User Insights 

  • Well-defined goals like restaurant names are handled accurately. However, searching for a dish name, like chimichanga, shows results that may not match the keyword
  • Cuisine searched are handled appropriately, but feeling requests like "spicy" or "breakfasty" are not handled well. 
  • Provides results as unstructured data without taking into consideration user's culinary preferences. Thus, a lot of time is spent scanning through all the results to figure out the best option. 
Mapping a User Flow
User Flow.png
Figure 3: Workflow of a customer who does not have the Yelp app installed on their phone.

Impact on Design

Recommendations 

  • Add a face to the recommendation algorithm to mimic the feeling users get when a friend recommends them a restaurant. 
  • Make it easy to contact their friend,"Yara", through messaging services like WhatsApp, text messages, and Facebook Messenger. Just as you would contact a real friend.
  • Create a recommendation algorithm to provide users with a short, weighted, and tailored list of restaurants can consider their culinary preferences and restrictions. 
  • Utilize sentiment analysis, a hybrid recommender system, and contextual and discriminatory models to personalize the experience for users. 
Impact on Design
Yara.png
Figure 4: Yara, the friendly face of our Intelligent Recommendation Algorithm

Integrating the AI

Flow Chart - AI.png
Figure 5: Intelligent Architecture Flow Chart I created
Integrating the AI

Prototyping

On-boarding Process

Prototyping
iteration 1.jpg
Figure 6: First Iteration of On-boarding Process
final iteration 1.png
Figure 7: Final Iteration of On-boarding Process
Feedback between first and final iteration: ​
  • The on-boarding process is too long
    • If it is too long, new user's will abandon
  • User's will not want to input explicit data to fine tune the algorithm upfront

Fine Tuning the Recommendation Algorithm ("Hanging out with Yara")

iteration1.2.png
Figure 8: First Iteration of Fine-tuning Process
final iteration.png
Figure 10: Final Fine-tuning Process and Results Design
Feedback between first and final iteration: ​
  • Iteration 1 is reminiscent of Tinder and will be tedious
  • Search results should clearly indicate searched word

Messaging Yara (the recommendation algorithm)

messaging iteration 1.png
Figure 11: First Iteration of "Messaging Yara"
final iteration 2.png
Figure 12: Final Iteration of "Messaging Yara"
Feedback between first and final iteration: ​
  • "Yara" should have more of a personality in her responses
  • Adding a friendly personality will make users more comfortable with responding back
bottom of page