Scribd
Upload
17 views

Recommendation System

Uploaded by

chetannanglot195
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
17 views

Recommendation System

Uploaded by

chetannanglot195
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
You are on page 1/ 21

Introduction to

Recommender
System
Understanding the Basics and
Applications
Presented by :
🎓 Sahyadri Chalak-AI2163
Problem statement
Develop a personalised
recommendation system for
an online retail platform to
enhance user experience ,the
system should analyze user
behavior..
What is Recommendation system

A recommendation system (or recommender system) is a type of software


tool or algorithm designed to suggest items, content, or services to users
based on various types of data. These systems are widely used in digital
platforms to personalize user experiences and help people discover new
content or products that align with their preferences.
Literature Survey
Introduction & Evolution
● 1990s: Early systems like Tapestry (1992) and GroupLens (1994).
● Growth: Collaborative Filtering (CF) & Content-Based Filtering (CBF).

Key Techniques
● Collaborative Filtering (CF):
○ User-based (Shardanand & Maes, 1995)
○ Item-based (Sarwar et al., 2001)
● Content-Based Filtering:
○ Document recommendation (Pazzani & Billsus, 1997)

Hybrid Systems
● Combines CF & CBF (Burke, 2002).
● Netflix’s success using hybrid models.
Matrix Factorization & Deep Learning
● Matrix Factorization: Netflix Prize (Koren et al., 2009).
● Deep Learning: YouTube model (Covington et al., 2016).
Challenges
● Cold Start Problem (Schein et al., 2002).
● Scalability & Sparsity.
Recent Trends
● Explainability (Zhang & Chen, 2020).
● Privacy-preserving systems (Canny, 2002).
References:
● Key papers and authors listed briefly.
Visuals/Design:
● Use icons for types of recommendation systems (CF, CBF, Hybrid).
● Simple flowchart or timeline to depict evolution.
● Graphical representation of recent trends (Explainability, Privacy).
Types of Recommender
Systems
o Collaborative
Filtering
o Content-Based
Filtering
o Hybrid
Approaches
Collaborative
Filtering
⮚ These systems make
recommendations based
on similar user
preferences and
behaviors, such as
purchases or ratings.
Content-Based Filtering

⮚ These systems analyze the


features and attributes of
items to make
recommendations based on
a user's past preferences.
Hybrid
Approaches
{Collaborative+Content based}
How Recommender Systems Work?

• User Data:
• Item Data:
Information about Attributes and
user preferences, metadata of
behavior, and items (e.g.,
interactions (e.g., genres of
clicks, purchases, movies,
ratings). categories of
• Interaction Data: products).
Data on how
users interact with
items (e.g., time
spent on a product
page, ratings
given).
Algorithms in Collaborative Filtering

User-Based Item-Based

• k-Nearest Neighbors (k-NN) • Item-Item Similarity Algorithm


(If a user likes a specific book, the system
Algorithm(e.g.If User A and User B have might recommend other books that are
similar tastes in movies, and User A likes a
similar based on ratings from users who
movie that User B hasn’t seen yet, that movie
also liked that book.)
might be recommended to User B.)
• Similarity •Similarity Measures for Items
(Adjusted Cosine Similarity(i,j)=∑(ru,i −rˉi )2 ⋅∑
Measures(Cosine Similarity(u,v)=∑ru,i2 ⋅∑r (ru,j −rˉj )2 ∑(ru,i −rˉi )⋅(ru,j −rˉj ) )
v,i2 ∑ru,i ⋅rv,i ,Pearson(u,v)=∑(ru,i −rˉu )2 ⋅∑(rv
,i −rˉv )2 ∑(ru,i −rˉu )⋅(rv,i −rˉv ) )
Algorithms in Content-Based Filtering

Decision Trees Naive Bayes Classifier Cosine Similarity Neural Networks


A probabilistic classifier
based on Bayes’
A decision support tool Cosine similarity
theorem with an Deep learning models
that uses a tree-like measures the cosine of
assumption of capable of learning
model of decisions and the angle between two
independence between complex patterns in
their possible vectors, representing
features. data.
consequences. their similarity.

Used to calculate the


Categorizes items similarity between a
Captures intricate
Can be used to classify based on their features user's profile and item
relationships between
items into categories and user’s preferences, features. Items with
item features and user
based on features and recommending items higher cosine similarity
preferences, providing
make recommendations that fall into categories scores are
sophisticated
based on these preferred by the user. recommended.
recommendations.
classifications. P(C∣X)=P(X)P(C)⋅P(X∣ Cosine Similarity(A,B)=
C) ∥A∥∥B∥A⋅B
Algorithms in Hybrid Recommender
systems
Weighted Hybrid Switching Hybrid

⮚ Combines multiple recommendation


techniques by weighting their outputs and ⮚ Selects between different recommendation
then aggregating them. algorithms based on certain criteria or user
context.

⮚ Recommendation Score=w1 ⋅ScoreCF +w2 ⋅


ScoreCB +⋯ where w1,w2,…w_1, w_2, \ ⮚ Choose the recommendation algorithm based
ldotsw1 ,w2 ,… are weights assigned to the on user context or system state (e.g., use
different recommendation techniques (e.g., collaborative filtering for users with rich history,
collaborative filtering, content-based and content-based filtering for new users)
filtering).
USE OF COS FUNCTION IN
Recommender System
1-cos similarity=cos distance

d
Cos similarity = cos θ

Cos distance = d
1 2
X1 X1

* P1

90

P2 P1 P2

* X2
* * X2

1-cos(90)=cos distance 1-cos(0)=cos distance


1-0=cos distance 1-1=cos distance
1= cos distance 0= cos distance

i.e It doesn’t Recommend i.e It will Recommend


Challenges in Recommender Systems
Applications of Recommender Systems
Future Trends in Recommender
Systems

1.Integration of Artificial Intelligence and Machine Learning


2. Personalization Beyond Traditional Data
3. Privacy-Preserving and Secure Recommendations
4. Cross-Domain and Multi-Modal Recommendations
5. Explainability and Transparency
Conclusion

The future of recommender systems is poised to leverage


advancements in AI, machine learning, privacy technology, and user
interaction to deliver more personalized, accurate, and secure
recommendations.
Thank you,

Sahyadri Chalak
SY AIML
DIEMS .

You might also like