Machine Learning

Recommender Systems
 Learning Objectives

By the end of this lesson, you will be able to:

Examine different use cases and delve into different recommender

systems and their designs

Build a recommender engine using PyTorch

Inspect different filtering techniques

Discuss dimensional reductions and matrix factorization
 Business Scenario

An e-commerce company is experiencing a decline in customer engagement and

sales due to outdated product recommendations. To address this issue, the
company has decided to implement a new recommendation engine that uses
machine learning algorithms like collaborative and content-based filtering, as well
as hybrid techniques.

The new engine will gather, analyze and filter data in real-time, serving a global
user base with advanced search and upsell capabilities. Personalized
recommendations will be provided to customers, enhancing the overall user
experience and increasing engagement.

The recommendation engine aims to automate the process of finding new

products that users are likely to enjoy, resulting in higher sales for the company.
By adopting a modern and effective approach to recommendation, the e-
commerce company aims to remain competitive and retain customers.
How Do Recommendation Engines Work?
Discussion: Recommender Systems

• What are recommender systems?

• What are the different types of recommender

systems?

• Give examples of recommender systems.

 Working of Recommendation Engines
1
A recommendation engine is a data filtering tool that uses machine learning algorithms to recommend
the most relevant items for a particular requirement.

There are generally three main types of recommender systems:

Collaborative filtering

Content-based filtering
2
Hybrid filtering
 Working of Recommendation Engines
1

Collaborative filtering: considers similarities

Example: Recommendation based on
between users and items to provide interests of similar users
recommendations

Content-based filtering: analyzes data to Example: Product recommendation based

create a list of recommended products on a user’s browsing history

2
Example: Netflix recommendations based
Hybrids filtering: benefits from the advantages
on users’ watch history as well as search
of the other two types
habits of similar viewers
 Recommendation Engines

Recommendation engines process data via the following phases:

1. Collecting: data such as 4. Filtering: data is filtered

page views, view history, or to narrow down on
cart events may be relevant information and
procured make recommendations

2. Storing: data is stored in 3. Analyzing: similar items

databases or object are collated and matched
storage
 Recommendation Engines

Analytical techniques may be implemented in real time (in seconds) or near real time (in minutes) and
may also be performed in batches.

Real-time systems are preferred when instant

recommendations are necessary.

Batch analysis may be performed for periodic

analysis requirements, such as daily sales.
 Application of Recommendation Engines

E-commerce sites and Social networking

streaming services platforms

Use product recommendation Use recommendation systems to

engines to customize what to connect users with similar
display to visitors based on their interests based on their online
preferences and browsing history activity

Internet service providers Banks

Analyze credit card usage and

Modulate offerings based on
account activity to recommend
clients’ internet consumption
relevant wealth management
patterns
products to customers
Use Cases of Recommendation Engines
 Use Cases of Recommendation Engines

From the news feed on our phones to e-commerce products, the work of recommender systems is
visible everywhere.

Let us investigate a few use cases.

 E-Commerce Sites

Recommendation engines are responsible for the personalization of products ranging from clothing to
offers on gadgets, jewelry designs and food.

This is done by collecting and analyzing data on:

Past purchases

Social media interactions

Content interactions
 Social Media Platforms

They leverage recommendation engines to:

Support common social behavior Target better customers

Target increased interactions Target better matches

It enables personalized product discovery by users and helps businesses showcase the right stock
keeping units (SKUs) to customers and customize discounts, plans and other offerings.
 Media Platforms

They use recommendation engines to show the most relevant content to users based on their history
and interactions.

Some of the platforms that use recommendation engines are:

 Search Engines

Each user receives unique search results that are tailored to them based on their past clicks and
activities.

Recommender systems leverage image, text

and even voice data to be prepared with more
relevant search results.
Examples of Recommender Systems and Their Designs
 Examples of Recommender Systems

Recommendation engines are used across various industries such as:

Finance companies Social networks

6
1

Internet service providers 5

2 Advertising engines

4
Retail outlets 3
Search engines
 Design Factors

The design of the recommender systems depends on:

Data parameters involved

Need for real-time or Need for recommendations to be user-

passive recommendations based, item-based, or content-based

Seasonality of data Frequency of feedback

Environments being online, offline, or

Periodicity of data
nearline
 Design Factors

While designing recommendation engines, one must consider the main purpose of the application.

Example

News portals may have some sections with the same

content for all based on the overall popularity, and
some sections customized for the local user.

The application can be for a global user base, hyperlocal, or highly customized.
 Examples of Recommender Systems

Let us examine the design of recommendation systems with some sector-specific use cases.

In retail, recommendation engines come with advanced search and upsell capabilities.
 Examples of Recommender Systems

Retailers like H&M recommend Others also bought products for users searching for fitted tops.
 Examples of Recommender Systems

After selecting a particular top, recommender systems also recommend the tops that others have
bought under Others also bought.
 Examples of Recommender Systems

Sites like Goodreads provide READERS ALSO ENJOYED and See similar books options.

There is a recommendation engine at work here.

 Examples of Recommender Systems

AI-based recommendation systems account for the hyper-personalized recommendations on

media sites.

Upsell features such as More Like This keep viewers hooked to such portals.
 Examples of Recommender Systems

Recommendations like Related to items you’ve viewed can be seen on e-commerce sites.

Constant feedback from users is essential for the engines to make relevant recommendations.
Discussion: Recommender Systems

• What are recommender systems?

Answer: A recommendation system is a subclass of Information
filtering Systems that seeks to predict the preference a user
might give for an item.

• What are the different types of recommender systems?

Answer: The different types of recommender systems are
collaborative filtering, content-based filtering, and hybrid
filtering.

• Give examples of recommender systems.

Answer: Recommendation engines are used across various
domains such as Youtube, E-commerce, OTT, and Book selling.
Leveraging PyTorch to Build a Recommendation Engine
Discussion: Pytorch

• What is PyTorch?

• Specify the benefits of PyTorch.

 PyTorch

PyTorch is an open-source library that connects research-oriented inventions and discoveries in

machine learning to commercial applications.

Example

It allows customization of algorithms on

mathematical findings to leverage these for
commercial applications.

One such application is recommendation engines.

 Leveraging PyTorch

Collaborative filtering is one of the methods used to build recommendation engines.

It performs automatic prediction based on the following principle:

If users A and B like item T1 and user B likes item T2, then user A might have a
higher chance of liking T2 based on preference similarity.
 Leveraging PyTorch

Collaborative filtering is used in movies, arts and location recommendations.

Predictions of user likability for different items can

be made, enabling users to view items they
haven't seen before.
 Leveraging PyTorch

Since the goal is to produce predictions for every combination, the data complexity becomes high.

Therefore, PyTorch coding offers algorithm optimization.

 Leveraging PyTorch

Since the goal is to produce predictions for every combination, the data complexity becomes high.

Therefore, PyTorch coding offers algorithm optimization.

 Types of Collaborative Filtering: Memory Based

It is a recommendation engine where user predictions are based on the item or user similarity.

The time-space complexity for all combinations in memory-based recommendation engines may be
very high if all combinations of item preferences are considered.
 Types of Collaborative Filtering: Model Based

Model-based collaborative filtering refines the principle of similarity-based recommendation

depending on the model used, such as PCA and clustering.
Types of Collaborative Filtering: Hybrid

It is a combination of the other two methods.

 Discussion: PyTorch

• What is PyTorch?
Answer: PyTorch is an open-source machine learning (ML)
framework based on the Python programming language and the
Torch library. It is used for developing and training neural network-
based deep learning models.

• Specify the benefits of PyTorch.

Answer: The benefits of PyTorch are:
• It offers developers an easy-to-learn, simple-to-code
structure that's based on Python.
• It enables easy debugging with popular Python tools.
• It offers scalability and is well-supported on major cloud
platforms.
• It provides a small community focused on open source.
Assisted Practices

Let’s understand the topics below using Jupyter Notebook.

• 8.06_Collaborative Filtering and Memory-Based Modeling

• 8.07_Item-Based Collaborative Filtering
• 8.08_User-Based Collaborative Filtering
• 8.09_Model-Based Collaborative Filtering
• 8.10_Dimensionality Reduction and Matrix Factorization
• 8.11_Accuracy Matrices in ML

Note: Please download the pdf files for each topics mentioned above from the Reference
Material section.
 Key Takeaways

Recommendation engines use machine learning algorithms to filter

and recommend relevant items.

There are three types of recommendation engines: collaborative

filtering, content-based filtering and hybrids.

Recommendation engines process data through the collecting,

storing, analyzing and filtering phases.

They have various applications including e-commerce, social media,

internet service providers and finance companies.

PyTorch is an open-source library used to build recommendation engines

based on user preferences and is commonly used in e-commerce, retail
and social media.
Knowledge Check
Knowledge
Check
What are the three types of recommendation engines?
1

A. Hybrid, analytical and batch

B. Collaborative filtering, content-based filtering and hybrid

C. Online, offline and nearline

D. Real time, passive and hybrid

 Knowledge
Check
What are the three types of recommendation engines?
1

A. Hybrid, analytical and batch

B. Collaborative filtering, content-based filtering and hybrid

C. Online, offline and nearline

D. Real time, passive and hybrid

The correct answer is B

There are three types of recommendation engines: collaborative filtering, content-based filtering and
hybrid.
Knowledge
Check
What is the purpose of data collection in recommendation engines?
2

A. To perform real-time analysis

B. To create a list of recommended products

C. To narrow down relevant information

D. To procure data such as page views, view history, or cart events

 Knowledge
Check
What is the purpose of data collection in recommendation engines?
2

A. To perform real-time analysis

B. To create a list of recommended products

C. To narrow down relevant information

D. To procure data such as page views, view history, or cart events

The correct answer is D

Data collection is the first step in the recommendation engine process, and it involves procuring data
such as page views, view history, or cart events.
Knowledge
Check
What is collaborative filtering?
3

A. Analyzing data to create a list of recommended products

B. Modulating offerings based on clients' internet consumption patterns

C. Providing recommendations based on the interests of similar users

D. Filtering data to narrow down on relevant information

 Knowledge
Check
What is collaborative filtering?
3

A. Analyzing data to create a list of recommended products

B. Modulating offerings based on clients' internet consumption patterns

C. Providing recommendations based on the interests of similar users

D. Filtering data to narrow down on relevant information

The correct answer is C

Collaborative filtering is a type of recommendation engine that provides recommendations based on

the interests of similar users.
Thank You!