Facebook Local Discovery
Connect customers and local businesses through Facebook Recommendations.
My role: Product Design Intern
Time: from May 2018 to August 2018
As a Product Design Intern at Facebook. I’m with the local discovery team working on Recommendations. During the 12 weeks internship, I executed design projects on building new interactive tools and experience to encourage people to leave meaningful and helpful recommendations. Provided ideas from various angles to improve users’ motivation of engagement.
Testimonials from peers:
“From the small sampling of work I’ve seen, she has a strong sense of hierarchy, color and layout.”
“She’s taken several different creative approaches to solving a hard problem that’s been persistent in the product for years.”
“I’ve been very impressed with how quickly and enthusiastically Yumeng moves to identify problems and areas of improvement within the product.”
Topic: Helpful Recommendation
Context: Small businesses owners across the world depend on word-of-mouth recommendations to grow their businesses. Our team helps them do just that by encouraging friends to share recommendations on Facebook.
Task: How might we encourage people to leave more helpful recommendations?
Problem: For many consumers, however, a “Yes or No” recommendation alone isn’t enough to make a decision about a business.
Need: They are looking for helpful content, such as details, specifics, and unique insights about a business.
There are couple things you can do with recommendation, you can ask a recommendation and post it on news feed, or you can comment to someone’s post to give a recommendation, and my focus specifically on page recommendation production, which means you can go to a business page, recommend or not recommend it, and write a review.
My journey span three phases, I spent 1–2 weeks to understand the recommendation product, and spent the most of time explore and refine.
Learn from Research:
The highlight from research result is: ratings and reviews are the most prominent consideration signals. Ratings, reviews, photos and comments on photos got most reviews, and those are the trusted sources, the word of mouth, that the information people can rely on to make a decision.
Based on the understanding, I came up three high level design hypothesis that I believe in that can make recommendation helpful
Assumption 1: Increase the precision and specification of signals can make recommendation more helpful.
More precise and specific consideration signals can give consumers better sense to understand a business to make decision. What are some signals we have now and how to improve it?
- Scores: our score is overly higher than the scores on other websites, if we can get more precise data on how strongly people recommend or not, it will benefit our score to be more accurate and credible.
- Binary Choice: from contributors’ perspective, YES and NO is not sophisticated enough for them to input precise and complex opinions. Creating new tools to collect more precise feedback will be helpful.
- Ratings and Review: from consumers’ perspective, the signal people get from YES and NO is not as strong as ratings. Based on our research result, ratings and reviews are the most prominent consideration signals. Instead of stars for rating, we can create new ways to bring more spice to YES and NO.
- What and Why: currently we are able to get strong signal on whether people recommend a business or not, but we are lack of the understanding of what and why people are recommending. Gaining more signals on “What” and “Why” would make the recommendation more relevant and reliable.
Exploration and Refine for Assumption 1
I spent a lot of time on implementing multiple ideas into one design flow, and I realized it’s really hard to move forward this way, because it’s hard to verify and test multiple ideas at same time.
So, I breakdown the design assumptions into multiple experiments. I learned that I need to start from the minimum thing I can do and develop one idea at one time.
Design Experiment 1
The topic for the first experiment is “how strongly do people recommend a business”, to reach the goal, I need to think of what is the minimum thing I can do here to collect the signal successfully and also fit in the current flow.
Here are some different versions of tools for contributors to input the signal. Compare to binary choice YES and NO, the purpose of collecting this signal is to enable granular data input from contributors to benefit our score to be more accurate and credible, and also allowing users to express their feelings sensitively.
- Data input: Slider, Long press, 3D touch, Trinary choice, etc.
- Communication: Expression words, Numbers, Percentage, etc.
Thinking from consumers’ perspective is also important, the design component for them would be the presentation of different levels of recommendation. We found that by using words, we can create a conversational rating system, which is more interesting than star ratings.
This is another iteration for trinary choices. I worked with a content strategist on my team to figure out what are some different levels of expression, and how many levels we want to offer users. The solution I got from them is three levels of expression: Highly recommend, Recommend, and Somewhat Recommend.
Final decision for Design Experiment 1:
The reason I chose slider is that it’s the best tool to enable granular input, and compare to other ideas I have, it’s more intuitive for people to understand and interact with.
The challenge of this design is there is a conflict between trinary expressions and granular slider. The iterations of the final design focused on how to make the three levels of expression and the slider most compatible.
Rolling Research Result: Over all 4/6 found it direct, easy, and useful, however, 3/4 asked for some numeric reference.
Design Experiment 2:
The experiment 2 is an expansion of experiment 1. When we enable users to somewhat recommend a business, what negative signals we can get from them?
Solution: Offering positive and negative tags at same time.
Design Experiment 3:
The experiment 3 is also an expansion of experiment 1. How can we get more signals on what part of the business people recommend or not? How can we get more complex signals on some mixed positive and negative recommendations?
Solution: Categorize tags into different aspects to associate tags with a clear context. It’ll not only help users to easier find the tags they would use, also will bring us more detailed datas.
Other Design Experiments:
Assumption 2: Increase the quantity and quality of signals from non-power users
Increase the amount of recommendation contributors from non-power users will help us to boost the amount of signals we can get to enable us to bring more dynamic signals to consumers.
- Small pieces of information versus long reviews: The amount of non-power users are much more than power users. There are so much valuable information we can collect from them, such as the best food to order, for what type of occasion, etc. Those are specific and helpful information that consumers are looking for. If we can increase the amount of contributors in general and try to get those small pieces of information from non-power users, we can bring more dynamic signals for consumers.
- Motivations: We are losing the opportunities to get feedback from non-power users because they are lack of motivations to leave recommendation. We should stimulate their existing potential motivations and create new motivations for them to leave recommendations
Design Experiment 1: Writing Prompt 1
Design Experiment 2: Writing Prompt 2
What are some good prompts could be?
Firstly, Business owners can provide the questions or topics that they want reviewers to talk about the most. The topic could be their most competitive strength and what people liked the most about them. Secondly, Collecting questions from how people ask for recommendations on their post. Look at what are some posts that get the most recommendation for this business.
Assumption 3: Matching consumers and contributors
Enable consumers to easily find the right signals that what they are looking for. It will make the recommendation more powerful for every single consumer based on different needs.
- Matching information providers and seekers: creating direct communication between consumers and contributors.
Design Experiment: Q & A Session
Q&A session aims to match information seekers and providers directly. Creating direct communication between consumers and contributors would enable consumers to easily find the right signals based on their specific needs. The business owner’s participation will also be helpful to build friendly customer relationships.