YelpViz, which is a visuaization system aiming at exploring restaurants locations with differnt health scores in major cities of US. This project of COMP4901F Data Visualization @HKUST and also a champion hackthon project for HealthCare division of hackUST.
It’s the first time that Yelp dataset provides photos for each restaurant, so we decide to make use of these images and hopefully we can have some new findings in the end. However, images provided by Yelp had no meaning since each image does not associate with its restaurant, thus they were just a mess originally. To deal with this challenge, we decided to crawl all food images in a given city and manually labled each image with a healthy score.
In this project, we used deep learning to predict how “healthy” a food is by analysing its photo as well as the title of the photo. First, after crawling more than 5000 food images of two major cities, Cleveland and Madison, from Yelp dataset and labling each image with a healthy score, we trained these images as samples with neural network GoogleNet, which achieved 85% accuracy in the end.
Then, the visualization system consists of five components - Control Panel, VizMap, CheckIn-Flow Clock, TextCloud, Review Sentiment Analysis Chart.
Control Panel: Enable city and restaurant selection, implement paging and healthy score sorting.
VizMap: Hexagon design to reduce visual clutters.
CheckIn-Flow Clock: Population flow indicates restaurant popularity and the time when people prefer having meals
TextCloud: Indicate food themes in the city and food popularity
Review Sentiment: customers' attitudes towards the given restaurant.
Note that the dataset is almost 2GB.
Python 2.7, virtualenv, node
# initalize virtual env
virtualenv venv
# activate virtual env
source ./venv/bin/activate
# install python dependencies
pip install -r requirements.txt
# run web server
python run.py# enter the front end sub-directory
cd ./client
# install front-end dependencies
npm install
# serve with hot reload at localhost:8080
npm run dev
# build for production with minification
npm run build#Install http-server
npm install -g http-server
# enter the front end sub-directory
cd ./client/images/
# start the server so we can access images
http-server -p 8888