AUDYOGIK SHIKSHAN MANDAL’S

Institute of Management & Computer Studies (MCA Institute)

(IMCOST)

MASTER OF COMPUTER APPLICATION (MCA)

UNIVERSITY OF MUMBAI

A.Y. 2021-22

Mini Project
On

Restaurant Recommender System

Group Member:

1.Siddhnath Gharge (MC2021031)

2.Saurabh Kamble (MC2021047)

3.Aman Sankhe (MC2021088)

4.Vishal Surve (MC2021109)

 Certificate
This is to certify that:
Group Member :1. Siddhnath Sunil Gharge

Roll Number: MC2021031

2. Saurabh Krishna Kamble

Roll Number: MC2021047

3. Aman Ashok Sankhe

Roll Number: MC2021088

4. Vishal Vilas Surve

Roll Number: MC2021109

of Master of Computer Application (MCA ) as undertaken & completed the Mini project work titled “
Restaurant Recommender System”during the academic year 2021-22 .
Submitted to this college in partial fulfilment of the curriculum of Master of Computer Application, University
of Mumbai.

Signature Signature

(Course Coordinator) ( Director IMCOST)

Date:
Place:

Seal
 Table of contents
Topic Page number
Introduction section 2
Data section 4
Methodology section 7
Result section 10
Discussion section 11
Conclusion section 12

Page 1
Restaurant recommender system

1. Introduction:
Restaurant recommender system is a machine learning model, developed to demonstrate
as a Summer Internship Project. It recommends restaurants based on user’s likes and
dislikes and his previous interest data.

Problem background:
Bangalore is the capital and largest city of the Indian state of Karnataka. With a
population of over 15 million (as of January 2016), Bangalore is the third largest city in
India and 27th largest city in the world.

The diversity of the cuisine available is reflective of the social and economic diversity of
Bangalore. Roadside vendors, tea stalls, South Indian, North Indian, Muslim food,
Chinese and Western fast food are all very popular in the city. Udupi restaurants, are very
popular and serve predominantly vegetarian cuisine. The Chinese food and the Thai food
served in most of the restaurants are can be customized to cater to the tastes of the Indian
population. Bangalore can also be called a foodie's paradise because of its vast variety of
foods and edibles with a touch of Bangalore's uniqueness and tradition.

Problem description:
Suppose I travel and keep changing places very frequently. This is very hectic and plus i
get to experience very different types of environment, of which I do not have much
knowledge about. In such situation, food can be an important factor for decided how you
rate your trips and plus also recommending it to the people. Food can also attract people
around to world to try it out if it were to be the best. In such scenarios, we need to find
the right place, at reasonable cost, to serve us the best possible way. So there are few
questions that must be addressed, such as:

1. How many types of foods are available in the restaurant?

2. Which is the most nearest to me with good rating?
3. How many "similar" restaurants are available nearby me?
4. Do the "similar" restaurants cost more? If so, what specialty do that have?

To address such question, XXYZ Company’s manager decides to allocate this project to
me not just to find out solutions to the questions but also build a system that can help in
recommending new places based on their rankings compared to the previously visited by
me.

Page 2
Expectations from this recommender system is to get answer for the questions, and in
such a way that it uncovers all the perspective of managing recommendations. It is
sighted to show:

1. What types of restaurants are present in a particular area?

2. Where are the similar restaurant present based on a preference to particular food?
3. How do different restaurants rank with respect to my preferences?

Target Audience:
Target audiences for this project does not limit to a person who keeps travelling but
everyone. People could simply decide to look for a similar restaurant all the time because
they are addicted to a specific category of food. People who rarely use restaurants would
prefer to have the most rated restaurants nearby them and all this could be easily handled
by our recommender system. So target for this project is basically everyone who is
exploring different places or similar places.

Success rate:
With restaurants evolving, new food categories emerge, hybrid food starts to be more
popular, we need a system that could help us access vast number of food varieties. It is
impossible for a person to ask each and every one about their visit to a particular place
and also not everyone remembers everything. On the other hand, Computers are good at
remembering things, and with Machine learning to its peak, it high time technology will
by our personal guidance and help us personally based on our likes and dislikes. So
people would care about this project as their personal assistance and success rate could
certainly increase with time.

Page 3
Restaurant recommender system

2. Data:
Data requirements:
To find a solution to the questions and build a recommender model, we need data and lots
of data. Data can answer question which are unimaginable and non-answerable by
humans because humans do not have the tendency to analyze such large dataset and
produce analytics to find a solution.

Let's consider the base scenario:

Suppose I want to find a restaurant, then logically, I need 3 things:

1. Its geographical coordinates (latitude and longitude) to find out where exactly it is
located.
2. Population of the neighborhood where the restaurant is located.
3. Average income of neighborhood to know how much is the restaurant worth.

Let’s take a closer look at each of these:

1. To access location of a restaurant, it’s Latitude and Longitude is to be known so that
we can point at its coordinates and create a map displaying all the restaurants with its
labels respectively.
2. Population of a neighborhood is very important factor in determining a restaurant's
growth and number of customers who turn up to eat. Logically, the more the population
of a neighborhood, the more people will be interested to walk openly into a restaurant and
less the population, a smaller number of people frequently visit a restaurant. Also, if more
people visit, better the restaurant is rated because it is accessed by different people with
different taste. Hence is very important factor.
3. Income of a neighborhood is also very important factor as population was. Income is
directly proportional to richness of a neighborhood. If people in a neighborhood earns
more than an average income, then it is very much possible that they will spend more
however not always true with very less probability. So, a restaurant assessment is
proportional to income of a neighborhood.

Data collection:
1. Collecting geographical coordinates is not difficult but after googling for more than 2
days, it was not available on open-source data websites such as Wikipedia, India gov
website, census report websites etc. So, I decided to use Google maps API to fetch
latitude and longitude but google API has limited number of calls that I could make with

Page 4
my free account. So, it would take around 15 - 20 days to fetch location of all the
neighborhoods in Bangalore.
Initially I scrapped list of neighbors using beautifulSoup4 from
[wikipedia](https://en.wikipedia.org/wiki/List_of_neighbourhoods_in_Bangalore). The
table headings becoming the boroughs and data becoming the neighborhoods. Bangalore
has 8 boroughs and 64 neighborhoods. So, I manually googled each neighborhood to find
its corresponding latitude and longitude. After doing so, I produced the following data
frame.

2. Population by neighborhood is again easy to find out given that it’s readily available.
But in case of Bangalore, it is again not the case. i was able to find population data for
few cities. [Here is the link] (https://indikosh.com/dist/655489/bangalore). Rest other
neighborhood population is assumed and may be inaccurate but since this is a
demonstrating project, the main idea to get the working model. The data frame for
Bangalore neighborhood population looks like:

3. Income by neighborhood is again easy to find out given that it’s readily available. But
in case of Bangalore, it is again not the case. i was able to find Income data for main city.
[Here is the
link](https://en.wikipedia.org/wiki/List_of_Indian_cities_by_GDP_per_capita).
Neighborhood Income is assumed and may be inaccurate but since this is a demonstrating

Page 5
Restaurant recommender system
project, the main idea to get the working model. The data frame for Bangalore
neighborhood population looks like:

4. Foursquare API:

Use of foursquare is focused to fetch nearest venue locations so that we can use them to
form a cluster. Foursquare API leverages the power of finding nearest venues in a radius
(in my case: 500mts) and also corresponding coordinates, venue location and names.
After calling, the following data frame is created:

Page 6
3. Methodology:
Exploratory analysis:
Scrapping the data from different sources and then combining it to form a single-ton
dataset is a difficult task. To do so, we need to explore the current state of dataset and
then list up all the features needed to be fetched.
Exploring the dataset is important because it gives you initial insights and may help you
to get partial idea of the answers that you are looking to find out from the data.
While exploring the dataset, I found out that Inderanagar has the greatest number of
venues while Varthur has the least.

Page 7
Restaurant recommender system
Also while producing graph for number of cluster, I produced a graph to explore all the
values for n_clusters and then finding the best by exploring the elbow graph.

Inferential analysis:
Most important factors while building the recommender system were population and
income. They are the most import factor because they have a nonlinear relationship
according to our dataset.
It needed to make some inferential analysis to understand this nonlinear relationship. As
the amount of population increases, it does not necessarily mean that average income of a

Page 8
neighborhood will also increase. It is true to most of the case but also many cases differ
to follow this trend. Similarly, a neighborhood with less number of people may not
necessarily have less average income. It is possible to have less number of people and
more income and vice versa. This can be inferred from the following graph:

Page 9
Restaurant recommender system

4. Result:
The result of the recommender system is that it produces a list of top restaurants and the
most common venue item that the user can enjoy. During the runtime of the model, a
simulation was done by taking ‘Whitefield’ as the neighborhood and then processed
through our model so that it could recommend neighborhoods with similar characters as
that of ‘Whitefield’.
The following image shows the result:

Page 10
5. Discussion:
Since there was a nonlinear relationship between income and population, it can be
concluded that we must always perform inferential approach to find relationship among
different set of features. Also during clustering, similar neighborhoods must be dumped
into the right cluster.
The following graph shows the clusters:

Another observation that we can make is that choosing number of clustering could
produce very diverse results. Some may be over fitted or some may be under fitted.
Hence analysis of number of clusters must be done. Ref elbow_graph in the Methodology
section.

Page 11
Restaurant recommender system

6. Conclusion:
The recommender system is a system that considers factors such as population, income
and makes use of Foursquare API to determine nearby venues. It is a powerful data
driven model whose efficiency may decrease with more data but accuracy will increase.
It will help users to finish their hunger by providing the best recommendation to fulfil all
their needs.

Page 12