The Recommendation Systems
Science Behind The Famous “You may also like…”
Ever wondered, what is the science behind the movie rating websites, online services recommending you the best products/books to purchase, providing the best search for your search on Google and the like? If yes, this post is for you.
Recommendation systems make use of Machine Learning techniques to provide the users with a list of recommendations based on the recommendation class opted. For example :
1. Movies, on the basis of ratings,
2. Books, on the basis of your search.
3. Products that you might like, based on the products purchased or searched for, in the past.
There are a number of recommendation system approaches available to design such a recommendation system, such as :
1. Stereotyping
2. Content-based Filtering
3. Collaborative Filtering
4. Co-occurrence recommendations
5. Graph Based
6. Global Relevance
7. Hybrid Recommendation Approaches
In this article, we will be talking about Content-Based Filtering, Collaborative Filtering and Hybrid Recommendation approaches which are the mostly used classes for building a recommendation system.
1. Collaborative Filtering
You remember that old-school approach of asking your friends about movies to watch, entertainment websites to visit, products to use/purchase?
This recommendation class works on the same principle and on a large scale i.e, number of people you would be considering for suggestions would be a lot more so as to ensure the best results.This is a user-based approach.
A collaborative filtering algorithm usually works by searching a large group of people and finding a smaller set with tastes similar to yours. It looks at other things they like and combines them to create a ranked list of suggestions. There are several different ways of deciding which people are similar and combining their choices to make a list. This list act as a list of recommendation for the users.
2. Content-Based Filtering
This is an item-based approach. The way the recommendation engine has been implemented so far, requires the use of all the rankings from every user in order to create a data set. It might work well like for a thousands of people or items, but a very large site like Amazon has millions of customers and products , and comparing a user with every other user and then comparing every product each user has rated can be very slow. Also, a site that sells millions of products may have very little overlap between people, which can make it difficult to decide which people are similar.
The algorithm begins by listing out the most similar items for each item. Then, when you wish to make recommendations to a user, you look at his top-rated items and create a weighted list of the items most similar to those.
3. Hybrid Recommendation Approach
This approach makes use of both the above mentioned recommendation classes to produce a list of recommendations to be provided to the user. This approach works better as it considers both the users as well as the items and hence producing a satisfying result based on that but whether to use it or not, totally depends on the recommendation system you’re designing.
Learnt something useful from the article? Kindly show your appreciation by giving a clap and sharing it with your friends. :)
