A Delightful Good Thing

A Delightful Good Thing

2017｜Type: WeChat Mini-Program｜Tag: Design Led｜Role: Product, UX Design

Background

The original intention of the demand side was to promote the Clean Plate Campaign through H5 WeChat Campaigns and complete the task of taking photos and checking in for five continuous days in this campaign to obtain the rewards. But I have doubts about H5 WeChat Campaigns and photo taking.

H5 WeChat Campaigns or WeChat Mini-Program

H5 WeChat Campaigns are characterized by timeliness and explosiveness in communication. With the continuous innovation of marketing campaigns and creative ideas, the spread of H5 will peak in a certain period, and then randomly decline rapidly. H5 obviously does not have the advantages of cultivating long-term habits.

Manual or machine learning

People record and praise food because of its beauty. This behavior occurs before a meal. When the user finishes the meal taking pictures and uploading them after the meal, the process is inevitably unpleasant, and it becomes unseemly and less than elegant. Users have difficulty in completing tasks. Perhaps we can shift the process of unpleasant images taken to machine, through machine learning image classifiers. The images are trained to classify and predict whether the plate is clean or not and how much food is left on the plate. User raw data is processed by the machine and fed back a delightful image to the user.

Therefore, we redefined the design challenge: to design a public welfare product - the Clean Plate Campaign - to cultivate employees' habits and awareness of cherishing food.

Problem

In the general public, dinner plates are immaculately white, but the reality is harsher. People will praise food and record it before the meal. But photographing it after the meal seems unseemly and not elegant enough. Users are challenged by both a lack of motivation and an aesthetic in their willingness to record.

Objective

Stimulate users to be willing to take pictures and keep do it.

Strategy

The indecency. The process is shifting the unpleasant images from manual to machine, which is trained to classify the images for food residue prediction.

The lack of motivation. The point reward is offered to those who finish each meal with a photo of a clean plate taken. Users can earn the point whether it’s positive or negative feedback.

Machine Learning

A total of 5,001 samples were collected over a three-week to artificially determine whether or not they were clean. The model has achieved 98.7% accuracy in the training phase, and user feedback will be introduced to train the model subsequently.

Design Challenges

The rejection of recognizing other items
The input image is too embarrassing
The data flow of machine learning
The Points and the Redemption
The first-time and non-first-time setting

Solution

Assist the user in finding the correct spot for the picture and ensure they upload the right photo. Communicate with the user through the CUI and display drawings instead of plain images.
If the result is a clean plate, you can earn 3 points for a clean plate and 1 point for feedback. Provide positive or negative feedback based on the accuracy of the prediction.

Clean Plate Flow

If the output is not a clean plate, 1 point will only be given when the user gives positive or negative feedback.

Non-clean Plate Flow

If the user puts in the same picture again, it'll be counted as invalid.

An Invalid Input

Use points to trade in to get people to engage with it frequently.
During the special month, double points could be activated.

Redeeming Points

Sure thing, the user can also check their own records. By seeing how much they’ve been involved, they can get into the habit of doing better.

Check the Record

Patent

Application: A method of image recognition and related equipment
Inventor: Zheng Yongsen; Zhu Xiaolong; Peng Xiang; Chen Xiaochang
Application number: CN201710844628.3
Application date: Sep 15, 2017

Effects

The unique users from Aug 7, 2017, to Dec 31, 2017, were 4,411 users. On World Food Day, the unique visitors were 994, the number of clean plates was 1,026, and the average daily clean plate rate was 82%, up 11%.

What’s next

Roadmap: Improve the impact.
Strategy: Record and save event tracking of actions in this WeChat Mini Program, and send the report of the Clean Plate Campaign.

Last but not least

This is the first time I've led the product design, and it's a little tiny project in terms of the users it serves. However, it is full of exploration of public welfare, experimentation of the new pattern, and new technology. It was this fearlessness that allowed a fresher to find his first confidence in the career of Technology for Good.

There was another challenge at the beginning, in January 2017, WeChat Mini Program was officially released, and there was no design experience to learn from. 2017 was also the year when AI started to take off, the new technology brought more design challenges, applying image training to product design, the ensuing recall, rejection, recognition, user feedback, and so on. Also, the scripting of CUI. All of this provided us with experience in designing for machine learning.