The recommendation system is prove to be very helpful in increasing revenue for big platforms like Netflix, YouTube, Amazon, Flipkart etc. Users are also introduced with items they like and they don’t have to rigorously search for it.
There is one method for recommendation system called Collaborative Filtering. It’s of two types :-
- User-based collaborative filtering
- Item-based collaborative filtering
User-based collaborative filtering
In this a user is recommended based on his/her similarities with other users. With an example everything will be clear. For example :- John like movie A and movie B. Sam watch movie A and liked then he will be recommended movie B. As John’s and Sam’s choices match that’s why the things one likes and other hasn’t tried will be recommend to him.
It is very computational expensive as number of users can be of large number. This system can be badly effected by creating fake account and spammers can ruin it.
That’s where comes our next system i.e Item-based collaborative filtering.
Item-based collaborative filtering
This is based on item-item similarity. We take items user has rated and on based of that we found similar items in the universe of items which user hasn’t rated. For example :- An item A is rated by two users, item B is not rated by any user, item C is rated by two users and item D is rated by one user only. Now if our user rate item C then item C will have 3 rating and similar to it is item A as it is rated by two users.
So like this a list of most similar item is prepared. This technique is better than User-based collaborative filtering because that is computational expensive and also the profile of users keep on changing therefore everything need to be computed again.