DATA ANALYTICS

UE22CS342AA2
UNIT-4
Lecture 56: Introduction to
Recommendation Systems

Gowri Srinivasa
Department of Computer Science and Engineering
 Data Analytics
Unit 4: Recommender Systems
Lecture 56: Introduction to Recommendation Systems (T2 Ch 1)

Teaching Assistant:
Ms. Amritha GK, PESU-25, Department of CSE
Slides excerpted from: “Recommender Systems:
The Textbook”, Charu C. Agarwal, Springer Slides collated by:
2016. Mr. Harshith Mohan Kumar, PESU-23, Department of CSE
[email protected]
Ms. Harshitha Srikanth ,VII Sem ,PESU, Department of CSE
Gowri Srinivasa [email protected]

Department of Computer Science and Engineering With grateful thanks for contribution of slides to:
Jyothi R, Professor at the Department of CSE, PESU
Unit IV – DA
Introduction
Outline:

○ Introduction
● Age of Recommendations
● Recommender Problem
● Example: Amazon

○ Goals of Recommender Systems

● Prediction vs Ranking
● Operational & Technical
● Applications

○ Basic Models

○ Types of Ratings

○ Challenges

○ Q&A
Unit IV – DA
Age of Recommendations

Bakhtiyari, Kaveh. “The Future Directions of Recommender Systems.” Medium, Futurist Zone, 4 Jan.
2021, https://medium.com/futurist-zone/the-future-directions-of-recommender-systems-72beae7c2dd2.
Unit IV – DA
Recommender Problem

❖ A Good recommender
• Show programming titles to a software engineer and baby toys to a new mother
• Don’t recommend items, which user already knows or would find anyway.
• Expand User’s taste without offending or annoying him/her..

❖ Challenges
• Huge amounts of data, tens of millions of customers and millions of distinct catalog
items.
• Results are required to be returned in real time.
• New customers have limited information.
• Customer data is volatile.
Unit IV – DA
Amazon a Personalized Store
Unit IV – DA
Amazon a Personalized Store
Unit IV – DA
Introduction Chapter (1.1)

● The basic idea of recommender systems is to utilize various sources of data to infer
customer interests.

● Usually the entity to which the recommendation is provided is referred to as the user,
and the product being recommended is also referred to as an item.

● These systems are used to predict the “rating” or “preference” that a user would give
to an item.

● The basic principle of recommendations is that significant dependencies exist

between user-and item-centric activity.

● Outline:
○ 1.2 - Main goals of recommender systems
○ 1.3 - Basic models & evaluation methods
○ 1.4 - Advanced models for recommender systems
Unit IV – DA
Goals of Recommender Systems (1.2)

Two primary models of the recommendation

problem
1. Prediction version of problem
○ The first approach is to predict the rating value for a user-item combination. It is
assumed that training data is available, indicating user preferences for items.

○ For m users and n items, this corresponds to an incomplete mxn matrix, where the
specified values are used for training.

2. Ranking version of problem

○ In practice, it is not necessary to predict the ratings of users for specific items in order
to make recommendations to users. The determination of the top-k items is more
common than the determination of top-k users.
Unit IV – DA
Prediction vs Ranking

Note: In the prediction

problem we would need
to get values for all
“node”
Xuan Huo, Ming Li,
On cost-effective software defect prediction: Classification or ranking?, Neurocomputing, Volume 363, 2019, Pages 339-350, ISSN 0925-2312, https://doi.org/10.1016/j.neucom.2019.05.100.
(https://www.sciencedirect.com/science/article/pii/S0925231219309221)
Unit IV – DA
Goals of Recommender Systems (1.2)
Operational & Technical Goals of Recommender Systems

1. Relevance: Users are more likely to consume items they find interesting, rating value for a
user-item combination.

2. Novelty: Recommender systems are truly helpful when the recommended item is something that
the user has not seen in the past. For example, Popular movies of a preferred genre would
rarely be novel to the user.

3. Serendipity: The items recommended are somewhat unexpected, and therefore there is a
modest element of lucky discovery. Recommendations are truly surprising to the user. It leads to
sales diversity or beginning a new trend of interest in the user.

4. Increasing Recommendation Diversity: It has the benefit of ensuring that the user does
not get bored by repeated recommendation of similar items.
Unit IV – DA
Spectrum of Recommendation Applications (1.2.1)

Popular historical & current

recommender systems

1. GroupLens Recommender System

○ Link to paper: (PDF) Recommender Systems: A GroupLens
Perspective

2. Amazon.com Recommender System

3. Netflix Movie Recommender System

4. Google News Personalization System

5. Facebook Friend Recommendations

Koidan, Kate. “How Netflix Uses Contextually-Aware Algorithms to Personalize Movie
Recommendations.” TOPBOTS, 11 Nov. 2020, https://www.topbots.com/netflix-movie-
recommender-system-rework/.
Unit IV – DA
Basic Models of Recommender Systems (1.3)

Collaborative Filtering Models

1. Memory-based (neighborhood-based): Ratings of user-item

combinations are predicted on the basis of their neighborhoods.
These neighborhoods can be defined in one of two ways:
i. User-based: The ratings provided by like-minded users
of a target user {A} are used in order to make the
recommendations for {A}.
ii. Item-based: Make the rating predictions for target item {B} Pinela, Carlos. “Recommender Systems - User-Based and Item-Based
by user {A} Collaborative Filtering.” Medium, Medium, 6 Nov. 2017,
https://medium.com/@cfpinela/recommender-systems-user-based-and-item-
based-collaborative-filtering-5d5f375a127f.

2. Model-based: Machine learning and data mining methods are used

in the context of predictive models.
Unit IV – DA
Basic Models of Recommender Systems (1.3)

More Recommender Systems

1. Content-based: The descriptive attributes of items are used to make

recommendations.
2. Knowledge-based: Recommendation process is performed on the basis of
similarities between customer requirements and item descriptions.
i. Constraint-based
ii. Case-based
3. Utility-Based: Utility function is defined on the product features in order to
compute the probability of a user liking the item.
Unit IV – DA
Basic Models of Recommender Systems (1.3)

Technique Background Input Process

Identify users in U similar to

Ratings from U of items in I
Collaborative Ratings from U of items in I u, and extrapolate from their
ratings of i

Generate a classifier that fits

Content-based Features of items in I U’s ratings of items in I U’s rating behavior and use it
on I

Identify users that are

Demographic information
Demographic information demographically similar to U,
Demographic about U and their ratings of
about U ad extrapolate from their
items in I
ratings of i

A Utility function over items

Apply the function to the
in I that describes U’s
Utility-based Features of items in I items and determine I’s rank
preferences.

Features of items in I.
A description of U’s needs or Infer a match between I and
Knowledge-based Knowledge of how these
interests. U’s need.
items meet a user’s needs.
Unit IV – DA
Types of Ratings (1.3.1.1)

Explicit vs Implicit
● Explicit
○ To collect explicit feedback from the user, the system must ask users to provide their
ratings for items
○ Requires direct participation from the user
○ Not easy to collect & doesn’t scale: only a small fraction of users leave ratings and
reviews

● Implicit
○ There is no user participation required to gather implicit feedback
○ System automatically tracks users’ preferences by learning ratings from user actions.
○ Ex: Which item they visited, where they clicked, which items they purchased, or how
long they stayed on a web page
○ Advantage of implicit feedback is that it reduces the cold start problems
Ahmad, Zahra. “Recommender Systems: Explicit Feedback, Implicit Feedback and Hybrid Feedback.” Medium, Analytics Vidhya, 23 Feb.
2021, https://medium.com/analytics-vidhya/recommender-systems-explicit-feedback-implicit-feedback-and-hybrid-feedback-
ddd1b2cdb3b.
Unit IV – DA
Domain-Specific Challenges (1.4)

Contextual Recommender Systems

1. Context-Based Recommender Systems (context-aware): Take various types of
contextual information (time, location, or social data) into account, while making recommendations.
2. Time-Sensitive Recommender Systems: recommendations for an item might evolve with
time.
i. The rating of an item might evolve with time, as community attitudes evolve and the interests
of users change over time. User interests, likes, dislikes, and fashions inevitably evolve with
time.
ii. The rating of an item might be dependent on the specific time of day, day of week, month, or
season.
3. Location-Based Recommender Systems:
i. User-specific locality: Geographical location of a user has an important role
ii. Item-specific locality: geographical location of an item (restaurant) might have an impact
4. Social Recommender Systems: based on network structures, social cues and tags, or a
combination of these various network aspects.
Unit IV – DA
Advanced Topics & Applications (1.5)

1. The Cold-Start Problem in Recommender Systems: Most people have

not rated most items
■ New items have no ratings
■ New users have no history
2. Attack-Resistant Recommender Systems
3. Group Recommender Systems
4. Multi-Criteria Recommender Systems
5. Active Learning in Recommender Systems
6. Privacy in Recommender Systems
7. Application Domains
Unit IV – DA
Q&A

1. Explain why unary ratings are significantly different from other

types of ratings in the design of recommender systems.

Unary ratings are mainly implicit feedback that we associate from a user to a given
item. Ones of the metrics I would use to evaluate such recommendation systems is Constraint-based Case-based
computing accuracy of estimated rankings either with utility-based measures or
with the receiver operating characteristic (ROC). Read More Specific requirements or Specific targets or cases are
constraints are specified.
specified by the user.
2. Discuss cases in which content-based recommendations will
not perform as well as ratings-based collaborative filtering. The original query is modified by The target is modified through user
addition, interaction, or the search results are
Content-based filtering can recommend a new item, but needs more data of user deletion, modification, or pruned
preference in order to incorporate best match. relaxation of the through the use of directional
original set of user requirements. critiques.
3. List the differences between Constraint-based and Case-based
Recommender systems.
Users are not in a position to Problem is addressed through a
exactly state conversational style of critiquing.
their requirements upfront.
4. Discuss a scenario in which location plays an important role in
the recommendation process.
Unit IV – DA
References

Text Book: “Recommender Systems: The Textbook”, Charu C. Agarwal,

Springer 2016.

Additional:
● “CS548S15 Showcase Web Mining.” Scribd, Scribd, https://www.scribd.com/presentation/414445910/CS548S15-Showcase-Web-Mining.

● Ahmad, Zahra. “Recommender Systems: Explicit Feedback, Implicit Feedback and Hybrid Feedback.” Medium, Analytics Vidhya, 23 Feb. 2021, https://medium.com/analytics-
vidhya/recommender-systems-explicit-feedback-implicit-feedback-and-hybrid-feedback-ddd1b2cdb3b.

● Wrg, Alexandre. “Image Recommendation Engine - Leverage Transfert Learning.” Medium, Towards Data Science, 8 Apr. 2021, https://towardsdatascience.com/image-
recommendation-engine-leverage-transfert-learning-ec9af32f5239.

● “Lecture 41 - Overview of Recommender Systems | Stanford University.” YouTube, YouTube, 13 Apr. 2016, https://www.youtube.com/watch?v=1JRrCEgiyHM.

● Koidan, Kate. “How Netflix Uses Contextually-Aware Algorithms to Personalize Movie Recommendations.” TOPBOTS, 11 Nov. 2020, https://www.topbots.com/netflix-movie-
recommender-system-rework/

● Pinela, Carlos. “Recommender Systems - User-Based and Item-Based Collaborative Filtering.” Medium, Medium, 6 Nov. 2017, https://medium.com/@cfpinela/recommender-
systems-user-based-and-item-based-collaborative-filtering-5d5f375a127f.

● “Recommender Extension.” e, http://elico.rapid-i.com/recommender-extension.html.

● Amazon.com. Spend Less. Smile More. https://www.amazon.com/.

THANK YOU

Gowri Srinivasa
Department of CSE, PES University
[email protected]