Proposal for Dish-Driven Restaurant Recommendations

To improve the ability of Yelp's search engine to suggest relevant restaurants based on a user's favourite dish. Currently, the results for a dish query favour highly-rated Chinese restaurants whose reviews might contain the individual keyword (e.g. 'noodles') but do not match the dish (e.g. 'ee-fu noodles'). Yelp publishes an updated academic dataset of business listings, anonymous user information, images and review texts for the Yelp Dataset Challenge - this is the data source for my project.

Please see the full report published on Rpubs: http://rpubs.com/eugenewoo/yelp_sbcapstone

Installation

Download Data

• Download the JSON files from Yelp into your working directory
  • You can find the data here.
  • You will have to register with Yelp before downloading.
  • For this project, you will only need to analyse the Business and Review files.
• Read the downloaded JSON files into R
  • Run stream_in from the 'jsonlite' package
  • Highly recommend to re-save in .Rds format using saveRDS; reading .Rds format using readRDS is significantly faster than reading JSON format and will make future sessions more efficient.

Package Requirements

• Install the following packages
  • dplyr
  • tidyr
  • ggplot2
  • jsonlite
  • stringr
  • qdap
  • scales
  • tm
  • SnowballC, followed by the statement update.packages("tm", checkBuilt = TRUE)

File Descriptions

• README.MD

• CODE

  • Yelp.R - Main R code for exploration, cleaning and text mining
  • Sentiment_scoring.R - Supplementary R code for sentiment analysis of restaurant reviews (*future work*)
  • server.R - R code for Shiny app backend to lookup restaurant names with the user inputs from ui.R
  • ui.R - R code for the Shiny app user interface that defines user inputs and page layout

• REPORTS

  • Mysteries of the Yelp Orient - Project Report.pdf - Full narrative containing analysis, findings and code snippets
  • Mysteries of the Yelp Orient - Presentation.pdf - Keynote deck summarising project