PatRec

Pattern Recognition Project 18794

Parameters to Run

		weights				knnK	numPrincipalComp

• Baseline [0, 0, 1] 10 10
• Vary W [1, 0, 0] 10 10 [.5, .5, 0] 10 10
• Vary K [.33, .34, .33] 10 10 [.33, .34, .33] 20 10 [.33, .34, .33] 30 10
• Vary PCA [.33, .34, .33] 10 20 [.33, .34, .33] 20 20 [.33, .34, .33] 30 20

Basic Outline

• 1. Data filtering and selection
  • Pull out the relevant users
  • Generate likeness factor (attractionFactor, digItFactor?)
• 2. Determine similar users
• 3. Predict establishment likeness factor for user
• 4. Recommend an establishment for the specified user

Current JSON Data

• Extracted elite users into _user_reduced_elite__training.json (20054 users) -- Can be used to train characterizing user models
• Extracted all restaurants into _business_reduced_restaurants.json (14303 restaurants)
• Extracted set of sample users (non-elite) into _user_reduced_sample__test.json (2014 users) -- Can be used for testing
• Extracted top 50 reviewers into _user_reduced_top50.json

TODO v2.0

• (UPDATED -- see below) Get user vector with restaurant types
• (DONE) Expand restaurant information with Price Range, Geograpich Information, Average stars (Baseline/Restaurant Prior) -- REMARK: Added to new file .._mapping_ext.csv for visual validation with old file -- Difference: Switched user_id and business_id columns while still sorted by restaurants
• (DONE) Predict Rating -> Only for selected user
• (DONE) Add prediction baseline where predicted user rating is just restaurant avg
• (DONE) Clean-up validation code
• (DONE) Incorporate PCA
• (DONE) Move knn and pca to a similar user class
• Analysis for report: Best K, Best User Features
• Add restuarant data to recommendation

Secondary goals:

• DigIt factor (norm. star=star/avg.star)
• Recommendation filter for nearby restaurants
• Implement further intelligent clustering/classification scheme with K-Means/Naïve Bayes/GMM..

Get user vector with restaurant types

• The user_feature_matrix script has been updated with current progress
• A description explains the issues involved
• Pseudo-code added as an example demonstration of how the user vectors could be assembled

Demo users:

Holly R. -- Dataset UsrID: Pv7DGHzZ-uqIUdOsqPpsVg

• Profile: http://www.yelp.com/user_details?userid=A2JGzkvNjckSmps_4FbKWw
• Review in Phoenix (top one): http://www.yelp.com/biz/little-miss-bbq-phoenix-2?hrid=K57-I9jZNV-2fWVG-JEdKw
• Taste: (Grill, BBQ, American)

Recommendation: http://www.yelp.com/biz/wicked-spoon-las-vegas

Terry T. -- Dataset UsrID: ZrUZtLgLOn15v2NIZl-oKQ

• Profile: http://www.yelp.com/user_details?userid=FcVQPI-ulgIXJcl5580YeA
• Reivew in Phoenix (top one): http://www.yelp.com/biz/treehouse-bakery-phoenix-2?hrid=jxkt7kpuTd0KZhabdDQ1wA
• Taste: (Desserts, Vegetarian, Bars, American)

Recommendation: http://www.yelp.com/biz/seasons-52-phoenix

Mia N. -- Dataset UsrID: wUEp87FqOB9ew27Iz6zk_w

• Profile: http://www.yelp.com/user_details?userid=Q6Dcxo3_lNCzkqK7J3FL7A
• Review in Phoenix (top one): http://www.yelp.com/biz/harumi-sushi-phoenix?hrid=UZBfZr_09HaypLGfzEm8LQ
• Taste: (Desserts, Bars, Sushi, Pizza)

Recommendation: http://www.yelp.no/biz/the-melting-pot-phoenix

Daren C. -- Dataset UsrID: C6IOtaaYdLIT5fWd7ZYIuA

• Profile: http://www.yelp.com/user_details?userid=XEHZoTDWjw3w-gcQyYIe4g
• Review in Phoenix (top one): http://www.yelp.com/biz/santos-lucha-libre-phoenix?hrid=h9GzwSM0NxjUq_WYrsQhWA
• Taste: (Mexican food, BBQ..)

Recommendation: http://www.yelp.no/biz/lucilles-smokehouse-bar-b-que-tempe