Recommendation systems

They are the systems that help people find things when process of finding the information you may need to make the choices is bit challenging due to many choices or maybe there are many alternatives. The essence of recommendation system is not new, even creatures like ants use recommendation to look for food. In the group of ants, each one of them goes to different direction to explore the space and finds out where the food is, then they return back leaving the trails. Then they let their community know about the food and everyone starts following the trails. Even cavemen used to follow the theory of recommendation system for e.g when a new plant grows outside their cave, they used to wait until someone eats them and check to see if that plant causes any harm to them. If it did not, then they started eating it.

All of the biggest online stores, applications and software use recommendation system today like Amazon, iTunes, NetFlix and many more. They create a magical store in which if a user visits their site, the whole store rearranges itself so that the store is personalized for that user and maximizes the chance of user buying the product. Online stores have no sales people to guide customers to find products they may buy like in physical store. Also online stores have million of products and there maybe products that are less famous but some group of people may buy them. Which means recommendation system solves the problem of discovery by recommending products according to user’s preferences. Recommendation system gives the type of system which rearranges itself according to user preferences.

For any online store, all they want is either revenue or engagement. If a store is selling products directly then they want user to buy as much as products they can and increase revenue like in Amazon. If it is a subscription service like NetFlix, they want users to spend as much time as they can.

One of the benefit of recommendation system is that user does not have to come up with lot of description about the product they are looking to find. The system may start to recommend contents or products once the user starts browsing similar stuffs.

Function of recommendation systems:

1. Filter relevant products and provide them to user
2. Predict rating the user may give to the products and sort the products according to rating in descending order
3. Predict if the user may buy the product
4. Rank products based on their relevance to the user

Some filtering approaches for relevant products

Here are some popular methods for recommending products to users.

1. Content Based Filtering

In this filtering process, we look into the attributes, descriptive characteristics and content of products and recommend other products that have similar properties. For example:

If Rabin searches or has purchase history for NVIDIA GeForce GT 610 graphics card, then content based filtering finds out the attributes for user would be “NVIDIA” and “Graphics Card”. Now we would want to show other NVIDIA graphics cards.

2. Collaborative Filtering

In this process, similarity is measured in terms of users. In content based filtering, similarity was measured in terms of content or attributes but here similarity is measured in terms of user’s behavior, purchase history and more. Here, users collaborate or help each other for finding out the products they may like. Lets take an example:

If person X likes books A and B

If person Y likes books A and B

If person Y likes book C

Then there is high probability that person X may like book C because from first two statements we knew that person X and Y are pretty similar.

3. Associative Rules

This type of filtering is mostly seen in cross sales in e-commerce sites. We recommend the user products that are associated to another product. In the first two types of filtering, they were based on similarity of products and users. But in this filtering, there is no similarity on product and users but we recommend the other products that are useful when buying some product. For example, if someone buys the android phone, there is high chance that he will also buy a cover for that phone so it would be wiser for us to show such products to users.

I am going to cover these filtering approaches in detail in coming posts. If you like the post, don’t forget to clap and follow me on medium and twitter.