ECE_682_Final_Project

Overview project current state (10 min)

We intend to recommend restaurants to individuals. Initially, our data source is the Kaggle Yelp dataset.

• Text parsing – predict stars, maybe if things get serious, predict health score

  • Approaches:

    • Surveying
    • NN -> needs a lot of prelabeled data (e.g. IMDB database?), but easier to implement (input review, label health code) (use SF kaggle dataset train)

• Cluster businesses by features -> predict rating

• Cluster users -> recommend restaurants liked by other users in the cluster

• Regress user reviews of restaurants -> predict restaurant scores

• Regress over community (30 closest users) on restaurant score(avg of 30 users) to predict score

• Generate random forests to predict restaurant score for a particular user.

Idea generation regarding information gathering (20 min)

How much of the data do we need to employ? I recommend just enough to get the job done – that is, start small and increment. Taylor said 50k user reviews and pair them with the health code, Julia said the SF health code dataset has ~50k

A second implementation of the NLP stuff is to predict some feature between the user and a restaurant. However, we require idea generation on the features to predict (response).

Which variables need to be in the user dataset? Which variables need to be in the business dataset?

Timeline + task breakdown (10 min)

Timeline

(1) must be done by Monday March 27th

(2) must be done by Monday April 3rd

(3) Must be done by Monday April 10th

(4) must be done by Thursday April 13th

(5) must be done by Wednesday April 19th

Tasks

• (1) Data wrangling

  • Build business dataset
  • Build user dataset

• (1) Text parsing

  • Build review -> health code dataset
  • (2) Star prediction
  • (2) Health Inspection Rating
• (2) Cluster by business
• (2) Cluster by users
• (2) Regress per user
• (2) Random forest
• (3) Regress for community
• (3.67) Analyze the model
• (4) Ensemble somehow
• (1-5) Report!!!!!!
  • Technical appendix
  • PR – 1page
  • FAQ – 1-4pages

Hand out the work (10 min)

FOR (1)

• Emma – Build business dataset
• Taylor – Text Parsing
• Julia – Build business dataset
• Dev – Text Parsing
• Andres – build user dataset
• Mac – Build user dataset
• Mac - will do all the random forest stuff.