pPreference-Driven
Restaurant Feed

As trust in star rating systems declined,
we created a personalized recommendation feed based
on user preferences.
This project revolutionized how people find restaurants
tailored to their unique tastes.

project background

Why Do We Focus on Preferences?

NAVER Corp., Korea’s leading search engine, is more than just a search tool; it’s a comprehensive lifestyle platform integrating news, shopping, maps, and more. It also holds the largest database of business information in South Korea, supported by a widely used star rating system.

However, this rating system began facing misuse, with false and malicious reviews harming small businesses and eroding customer trust, becoming a serious social issue.

To address this, NAVER introduced two key initiatives: launching Visit-verified reviews, validated through AI receipt recognition or card payment records, and developing a new evaluation system to replace outdated star ratings.

As part of the team, I helped develop this new system, creating a more personalized and accurate way to reflect diverse user experiences and highlight the unique qualities of each place, ensuring they’re discovered by those who will truly value them.

Replacing a limited, single-score system with personalized, user-specific preferences.

mission

The shift to preference-based recommendations required two key tasks: collecting individual preference data and developing a platform for personalized recommendations. This project focused on creating a Preference-Based Feed.

Our mission was to:

• Boost customer retention through recommendations, driving offline business growth.

• Restore trust in our review systems.

• Increase engagement with review content.​
  

my role

I was part of a cross-functional task force of about 30 people working on the design of a preference-based feed project from January 2020 to January 2021. The project involved collaboration with various teams, including map, AI and feed development. I contributed to UX research and data analysis, providing insights for project improvements, and participated in the screen design process.

Design solution 

Check Out Your Personalized
Restaurant Feed

Discover places perfectly suited to your taste, recommended based on your personal data like saved places, visits, and payment data.

Explore Reviewers
Who Share Your Tastes

Follow like-minded reviewers and
discover new dining spots that match
your preferences.

Focus on Reviews
from Your Favorite Areas

Add your go-to neighborhoods or dream destinations to get reviews centered around those spots.

Curious What Others Think
of This Restaurant?

Explore reviews from new users and broaden your culinary experiences.

How to got there 

To design the platform right from the start, we reviewed Naver’s existing recommendation services and analyzed competitors. We surveyed 40,000 users about their restaurant search habits and future service expectations. We also conducted in-depth video interviews with ten volunteers to gain deeper insights into user needs.

insight

During our user research, we explored the entire journey, yielding significant user insights.

Visual Information Preferred

While Naver has been known for text-heavy blog reviews, there's been a notable migration to image-centric platforms like Instagram among younger users, particularly when searching for restaurants. 

Ad Fatigue

The difficulty in distinguishing between genuine purchases and advertisements has led users to lose faith in reviews. Many users reported spending significant time and energy trying to determine whether reviews were authentic.

Trusted Endorsements Matter

Users tend to trust and follow recommendations from influencers they admire. Endorsements from influencers perceived as authentic and knowledgeable have a significant impact on consumer purchasing decisions.

Preference for Local Content

In O2O services, users prioritize relevance and proximity. Users preferr local, up-to-date, user-generated content over professionally edited material. 

We decided to create
a personalized recommendation feed

We designed a service using AI recommendation models to deliver personalized feeds, aiming to elevate reviews from simple ratings to high-quality local content.

Based on user-generated reviews, the feed promotes sharing preferences. Users can follow reviewers, explore their recommendations, and discover new places, enhancing their experiences.​

This service ultimately empowers users to manage their feeds actively, connect with like-minded individuals, and enjoy richer local experiences.

Design Exploration

We began exploring how to translate these insights into design solutions.

1. Focusing on the Image

Our prior research confirmed a need for image-focused content consumption.
We enlarged the image area and paired it with concise text to allow users to review visuals and text simultaneously.

To ensure quality while using visitor-uploaded images instead of promotional photos, we used an AI model to filter and maintain high standards without losing authenticity.

2. Ensuring Reviews Don't Resemble Advertisements

We decided to display only verified reviews from actual visits to prevent fake reviews. However, we realized that we needed to go further to enhance trust in the reviews.

In particular, we found that users might mistake recommendations for ads if the reasoning behind them isn’t clear. To address this, we implemented a feature that clearly explains why a recommendation is being made, emphasizing that it’s based on the user’s preferences, not business needs. This ensures users trust that the recommendations are truly personalized to their tastes.

3. Local Settings Configuration

In line with user feedback that knowing whether a recommended restaurant is accessible and locally situated is crucial, we implemented region-based filtering.

We designed the system to first inform users of their current location.
Additionally, in the settings, users can add preferred areas to customize searches within specific regions. For those unsure of which areas to explore, we also offer recommendations based on trending locations and the user's own search and visit history, enhancing their local discovery experience.

4. Expanding Tastes

Initially, we will recommend places and users similar to collected tastes. However, we wanted to go beyond this one-dimensional approach.
We advanced to the next level by including places visited by users with similar tastes, areas explored by followed reviewers, and thematic lists featuring their favored eateries.
We reasoned that if someone likes a place, others who like similar places will probably enjoy it too. This strategy extends individual taste horizons through social connections, fostering a community-driven expansion of preferences.

Optimizing User Experience:
Pre-Launch Usability Testing

The Naver app, visited by 30 million users daily, necessitated usability testing before deployment to ensure there were no usability issues. We conducted a beta test with users who had signed up for open notifications.

• Tab Enhancement
  • Expanded Regional Access: 
Users requested info outside their interest areas, so we added an 'All Regions' tab for simplified browsing.
     
  • Streamlined Following: 
Feedback prompted us to create a 'Followed Users' tab, focusing solely on content from followed users, aligning with user demand for targeted feeds.

• Streamlining Recommendations
  We streamlined the recommendation features to enhance the relevance and engagement of the content. This involved removing less popular recommendations and adjusting the visibility of others based on user engagement and feedback.
  
  
• Identified the need for broader variety of places
  We discovered a strong user demand for recommendations beyond restaurants, such as tourist attractions, shopping malls, and other points of interest. While this was part of our original plan, the clear user interest led us to prioritize adding this feature in future updates.

This approach allowed us to uncover user patterns and feedback that we hadn't anticipated, enabling us to address these issues before the official launch.

impact

I filed a patent for collecting and analyzing preferences to recommend places and users based on those preferences!

81% said it’s better than the previous system
1,000,000,000 reviews generated in 5 months
1,000,000 followers gained within 6 months
10,000,000 active reviewers reached in 3 years

Reflection

In the course of this project, I was concerned that preference-based recommendations might limit users’ ability to discover new content.
To address this, I aimed to incorporate diversity and serendipity into the system, much like how Netflix introduces unexpected movie suggestions to broaden user interests.

The ongoing challenge is to refine the system through continuous improvements, helping users discover places that resonate with their unique tastes, moving beyond simple ratings to highlight the specific qualities that make each location appealing to different individuals.