Date of Inspection Violations

Time Series Date of Inspection Violations

DBA’s with Most Violations

NYC Street Map

Zip Code

UBCF & IBCF Collaborative Recommendation Models

The UBCF and IBCF collaborative models were used to generate the Food Score Recommender system. The following is a breakdown of both approaches including validation metrics from both models. The UBCF: This method produces recommendations based on user-based collaborative filtering. The IBCF: This method produces recommendations based on item-based collaborative filtering.

[1] "realRatingMatrix"
attr(,"package")
[1] "recommenderlab"

[1] "POPULAR"

[1] 11268

$`'CESCA`
[1] 2.802913

$`'ESSEN`
[1] 5.678246

$`'RITAS`
[1] 2.802913

$`'WICHCRAFT`
[1] 4.802913

$`#1 Chinese Restaurant`
[1] 3.148508

$`#1 GARDEN CHINESE RESTAURANT`
[1] 2.802913

$`#1 NATURAL JUICE BAR`
[1] 4.136246

$`#1 SABOR LATINO RESTAURANT`
[1] 2.672806

$`$1 PIZZA`
[1] 2.802913

$`$1.25 PIZZA`
[1] 4.802913

$`% SHAO BIN ZHENG`
[1] 3.981842

$`& PIZZA`
[1] 3.678246

$`&PIZZA`
[1] 2.802913

 [1] "0"  "2"  "3"  "4"  "5"  "6"  "7"  "8"  "9"  "10" "11" "12" "13" "17" "21"

$`'CESCA`
[1] 13

$`'ESSEN`
[1] 11

$`'RITAS`
[1] 13

$`'WICHCRAFT`
[1] 13

$`#1 Chinese Restaurant`
[1] 12

$`#1 GARDEN CHINESE RESTAURANT`
[1] 13

$`#1 NATURAL JUICE BAR`
[1] 13

$`#1 SABOR LATINO RESTAURANT`
[1] 4

$`$1 PIZZA`
[1] 13

$`$1.25 PIZZA`
[1] 13

$`% SHAO BIN ZHENG`
[1] 12

$`& PIZZA`
[1] 11

$`&PIZZA`
[1] 13

IBCF Collaborative Model

Collaborative Model Creation

The model created is called IBCF or Item Based Collaborative Filtering trained with 5000 user ratings. The IBCF is similar to the UBCF or User Based Collaborative Filtering but allows for a less memory intensive approach without having to load the entire user database in memory, locally.

Essentially, this model computes internally the cosine similarity between all users represented as vectors, which in R is as simple as:

crossprod(a,b)/sqrt(crossprod(a)*crossprod(b))

Rec.model <- Recommender(dba_scores_grade[1:5000], method = "IBCF")

recommended.items.2 <- predict(Rec.model, dba_scores_grade[2], n=5)

# Display Top 5 Scores for 'ESSEN'
as(recommended.items.2, "list")

$`'ESSEN`
[1] "0"  "2"  "4"  "11" "5" 

recommended.items.2.top3 <- bestN(recommended.items.2, n = 3)

# Display Top 3
as(recommended.items.2.top3, "list")

$`'ESSEN`
[1] "0" "2" "4"

e <- evaluationScheme(dba_scores_grade[1:1000], method="split", train=0.9, given=1)

# Creation of recommender model based on IBFC
Rec.ibcf <- Recommender(getData(e, "train"), "IBCF")

# Predictions on the test data set
p.ibcf <- predict(Rec.ibcf, getData(e, "known"), type="ratings")

# Obtaining the error metrics 
error.ibcf <- calcPredictionAccuracy(p.ibcf, getData(e, "unknown"))

IBCF Model Results

error.ibcf

    RMSE      MSE      MAE 
1.970775 3.883953 1.811243 

User-User Collaborative Filtering

#In user-based collaborative filtering (UBCF) the procedure is to first find other users that are similar to a given user, then find the top-rated items purchased by those users. Those items are then recommended for the given user.

vector_ratings <- as.vector(dba_scores_grade@data)

kable(table(vector_ratings),  caption="Rating frequency")

vector_ratings = vector_ratings[vector_ratings != 0]
ratings = dba_scores_grade[rowCounts(dba_scores_grade) > 2, colCounts(dba_scores_grade) > 1]

dim(ratings)

[1] 5425   12

Normalize Ratings

ratings.n = normalize(ratings)
ratings.n.vec = as.vector(ratings.n@data)
ratings.n.vec = ratings.n.vec[ratings.n.vec != 0]

Splitting Data

For splitting data into test and train sets, we can use the evaluationScheme() function in recommenderlab. It extends the usage of generic methods of splitting the data, by allowing several parameters that are specific to recommender systems. As shown in the code section below, there is a parameter specifying how many items to use for each user, and another parameter specifying the minimum value that indicates a good rating.

set.seed(21)
percent_train = 0.8
items_to_keep = 1        # items to use for each user
rating_threshold = 1      # good rating implies >=1
n_eval = 1                

eval_sets = evaluationScheme(data = ratings, method = "split",
                             train = percent_train, 
                             given = items_to_keep,
                             goodRating = rating_threshold, 
                             k = n_eval)
eval_sets

Evaluation scheme with 1 items given
Method: 'split' with 1 run(s).
Training set proportion: 0.800
Good ratings: >=1.000000
Data set: 5425 x 12 rating matrix of class 'realRatingMatrix' with 19345 ratings.

We now build a UBCF model using the default parameters of the Recommender() function, and use it to predict using the test portion of the data set. We use library functions to evaluate accuracy of the prediction by comparing against values in the data set. Performance metrics for the UBCF model are displayed.

eval_recommender = Recommender(data = getData(eval_sets, "train"),
                               method = "UBCF", parameter = NULL)
items_to_recommend = 10
eval_prediction = predict(object = eval_recommender,
                          newdata = getData(eval_sets, "known"),
                          n = items_to_recommend,
                          type = "ratings")

eval_accuracy = calcPredictionAccuracy(x = eval_prediction,
                                       data = getData(eval_sets, "unknown"),
                                       byUser = TRUE)

UBCF Model Results

head(eval_accuracy)

                                          RMSE      MSE       MAE
'ESSEN                                1.986125 3.944691 1.5111111
#1 Chinese Restaurant                 2.128673 4.531250 2.1250000
1 OR 8                                2.173067 4.722222 2.1666667
104-01 FOSTER AVENUE COFFEE SHOP(UPS) 1.166424 1.360544 0.7619048
118 KITCHEN                           1.056531 1.116259 0.8714286
16TH AVENUE GLATT                     0.728869 0.531250 0.6250000

© 2020. The Data Incubator. All Rights Reserved.

Introduction

Data

NYC Restaurant Inspection Violations

The data used for Food Score was exclusively from NYC OpenData website, in particular the restaurant inspection results from August 2014 until present.

The NYC Inspection Violations can be found here:

• DOHMH New York City Restaurant Inspection Results

Problem

To put food safety and foodborne illness outbreaks into perspective, and why it’s a problem.

• Food safety is a concern in NYC as the number of violations have significantly risen by an average of 28% over the past 3 years.

• Across the USA there are 3,000 deaths a year due to food poisoning and foodborne illness outbreaks are common in NYC.

• New York City has one of the highest concentration of restaurants in the world with 27,000.

• Tourist revenue accounted for $44 billion in 2017.

Value Statement

Food safety is the central focus of the platform for consumers, tourists, or food and drink establishment revenue. Food Scores value statement is designed to drive value through tourist revenue and customer centric approach.

• The main feature of Food Score is building a recommendation system that predicts companies based on no critical flags, scores of less than 20, and only A scores.

• With 65 million people that traveled to NYC in 2018, accounting for $9 billion spent in food and drink establishments.

• The main deliverable is building a dashboard that integrates the recommender model that produces a list of the cleanest restaurants in New York City by borough and maps it.

NYC Competition

• SALT lets you view a restaurant’s menu, make a reservation through OpenTable when applicable, and even allows you to request an Uber to any of your saved locations.

• ChefsFeed capitalizes on the credibility and clout of leading professional chefs to help New Yorkers discover new spots in a social media-type network.

• PopCity allows users to map any food photos they find on social media outlets like Instagram or on the Popcity discovery channel. Using Instagram’s photo copy link feature, you can immediately import a post to your Popcity map.

End-Users

The end-user for Food Score is anyone looking to grab a bite to eat, whether it’s close to work, hotel, shopping, or most importantly close to Broadway in New York City. The value to the end-users would be in the form of knowing the cleanest/safest restaurants, their locations whether its by building, street, borough, or by cuisine type.

More specifically, end-users are:

• Restauranteurs
• Foodies
• Bloggers
• General Consumers

Summary

Food Score was created to provide a recommendation of the cleanest restaurants in New York City by borough. The recommendation system used was Recommenderlab’s UBCF/IBCF collaborative model. The recommendations model considered companies with only grade of A, score lower than 20, and no critical flags. There is a vibrant market for Food Score in NYC with similar competitors such as ChefsFeed, PopCity, and SALT. in NYC with similar applications.

Food Score provides a new unique approach to providing clean restaurants in NYC, and could impact:

• The NYC food scene.
• Reducing the likelihood of the next foodborne illness outbreak.
• Increasing tourist spending in restaurants by 1% would account for $90m in revenue.

Food safety and providing awareness of food safety is the core value and mission of Food Score. By recommending the cleanest restaurants in NYC, we hope to cut down on food poisoning deaths and foodborne illness outbreaks.

© 2020. The Data Incubator. All Rights Reserved.