1. Sriram Patil (201305532)
Nishit Soni (201002026)
Jiten Goyal (201101040)
Information Retrieval and Extraction (CSE474)
International Institute ofTechnology Hyderabad
Online Social Network Based Object
Recommendation System
3. Concept & Architecture
Social network based object recommendation
Recommending movies
Social Network: Facebook
Dataset: International Movie Database (IMDB)
Website
Login with
Facebook and
fetch users liked
movies
Server
Spark Web
Framework
Jetty Web Server
MySQL
IMDB Dataset
4. Challenges
Two different sources (Facebook and IMDB)
Sparsity
Even active users may have liked well under 5% percent of the
movies.
Scalability
Billions of users and Millions of movies.
Duplication
As movie ids are different. Have to match the movies with
names.
5. Problems with usual techniques
Nearest Neighbour algorithms require computation that
grows with both the number of users and the number of
movies. With billions of users and millions of movies, a
typical web based recommender system running existing
algorithms will suffer serious scalability problems.
Because of sparsity, a recommender system based on nearest
neighbour algorithm may be unable to make any movie
recommendation for a particular user. As, a result the
accuracy of the recommendations may be poor.
Ever growing data and users set.
6. Our approach
Fetch friends with whom the user has atleast some common
movies. If no common movies, then select all the friends.
Get movies liked by the friends.
For each movie, we get a recommendation score.
Parameters considered while assigning score:
Friends which share some likes with the user
Friends of same gender
Friends of same age group
Movies with same genre
Sort the score and return top “n” movies.
7. Recommendation relevance criteria
As the movies are recommended from a lot of friends, it is
little tricky to figure out if the recommendations are
relevant.
We used two criteria
Movies suggested by Facebook.
Our results are comparable to Facebook movie suggestions. And even
better in some cases.
It is a good recommendation if the user has already watched
that movie.
8. Enhancements
There are some more parameters which can be considered
while ranking a movie
Friend list like “Close Friends”, “Relatives”, etc can be given a
little extra weight.
Actors and directors of the movies can be considered when
ranking.
Similar recommendation systems can be extended to
recommend music, books, etc.