An introduction to Recommendation engines
and how these systems work.
Both content based and collaborative filtering models are introduced.
Hotel recommendation system is explained as a case study.
4. 4
Age group specific
Gender Specific
Region specific
Popularity based
Examples?
Non-personalized
RECOMMENDATION
SYSTEMS
5. 5
Why are they required?
Suppose person X likes Machine
learning, Data Science but
majority of the people like
Cricket. Will Non-personalized
recommenders be useful?
Personalized
RECOMMENDATION
SYSTEMS
7. 7
TFIDF algorithm (Term
Frequency*Inverse Document
Frequency)
Sample User Profiles:
News Recommendations
User profiling for
content based RS
1. Likings
Sports
Entertainment
Crime
2. Dislikings
Politics
International
8. 8
1. Titanic – Liked
Romance – 0.5
Adventure – 0.3
Drama – 0.15
Other - 0.05
2. Avengers – Disliked
Sci-fi – 0.3
Superhero – 0.4
Action – 0.2
Other – 0.1
Movie Recommendations
9. 9
User-User Similarity
Suppose person X likes products
A,B,C and person Y liked
products A,B,D.
Thus, the system will
recommend C to X and D to
Y.
Collaborative
filtering
10. 10
Hotel Recommender System – A Hybrid
Approach
1. Generate/gather a dataset with features required
2. Take required inputs from user (The more the better !)
3. Classify hotels from the whole dataset according to
user’s input
4. The Classified data is dealt with in two different methods
a) Non personalized approach
b) Personalized approach
5. In Non-Personalized approach the user is a first time
user
6. Personalized approach considers regular users because
their profiles are needed to be built.
12. 12
Collaborative Filtering Approach
1. Build profiles for every user
2. Correlate the required profile with others
3. The nearest neighbour or most correlated user will be
similar to the target user
4. Consider hotels booked or reviewed or liked by these
users to add bias to these hotels in classified hotel
list for target user
5. Direct collaboration would be really difficult
6. Hybrid approach using bias is feasible