Impact of Recommender System in E-Commerce



Did you realize that once you completing watching your favorite movie in Netflix, a list of other movies will be appeared in the recommendation section? And most of the time, you will be tempted or attracted to watch some of these recommended movies.


These are not some kind of random recommendation. What make this thing possible?


Let refresh some history here. Back in 2006, Netflix have announced "The Netflix Prize", a competition that was held with the intention to improve the efficiency of the engine by 10%. In 2009, Netflix have awarded $1 Million for developer who manage to increase the accuracy of the recommendation algorithm by 10%, and BellKor's Pragmatic Chaos has been announced as the winner of Netflix Price. The impact of 10% improvement of Netflix's recommendation algorithm has generates an average of 30 billion predictions per day. With this improvement, it has significantly increase customers satisfaction level, which could be indicated by reduction of churn data.


Meanwhile, another giant e-commerce company, Amazon also implement recommendation engine into their website, since 1998, using item-to-item collaborative filtering algorithm. If you are an Amazon's fan, you will realize that you become their fan because their system able to give good recommendation of what product that similar or might gain your interest. As reported by McKinsey, about 35% of consumers purchases on Amazon come from recommendations.


So, what is recommender system?


The goal of recommender system is to predict user's behavior and proposed appropriate product that they might interested to further purchase. It is one of the most popular machine learning algorithms used by e-commerce retailers. They are three distinguish type of algorithm used in recommender system, which are:


  • Collaborative Filtering Algorithm - recognizing pattern based on user interactions to filter product or item of interest for recommendation.

  • Content-Based Algorithm - suggest recommendation based on user's item & profile features.

  • Hybrid Algorithm - combination of Collaborative Filtering & Content-Based system.


Collaborative Filtering

Collaborative filtering is an algorithm used by giant industry such as Neflix and Spotify. It is a machine learning technique that learn the relationship between product and product/item. By this relationship, it will give recommendation of a product that their customer may like or enjoy.


The algorithm will learn through the historical data in the form of feedback matrix based on the rating by user on the item, and come out with a model that predict similarities of the user to provide recommendation. Report says that about 75% of item that people watch are from recommendation by Netflix. The algorithm can be divided into two popular approach:


  • User-Item Filtering - Users who has close similarity rating on some item may have similar preference on what they like.

  • Item-Item Filtering - Use the similarity of different item based on preference of users and make recommendation to another user who may interested with it.


Content-Based Algorithm

Content-based algorithm is most applied algorithm in e-commerce, and use by giant company such as Amazon and Aliexpress. The algorithm learns what item the user like and purchase, and provide recommendation on other similar or related items.

For example, when we ready to purchase keyboard and add it into cart, sometime the pop-up menu appears suggesting us to purchase mouse together with the message "You may like this item too". More advance algorithm will give you multiple selection with variety of price and feature that may attract you to purchase together with your first item.


KNIME as Visual Programming tools for Recommender System

KNIME is an open source visual programming software that apply drag and drop approach instead of writing programming script. In the next blogpost, I will discuss about the hands-on practice of constructing a recommender system using KNIME. Stay tune...


by Dr. Nickholas

Data Scientist & Trainer

107 views0 comments

Recent Posts

See All