Recommendation engines: a new effective marketing tool based on Machine Learning.


One of the latest applications of machine learning are recommendation engines. Businesses such as Google, Amazon and Netflix make use of recommendation engines in order to improve shopping experience and increase sales. Nowadays, customers quickly want to see what they like in order to fulfil their needs. However, lots of websites and services do not have a recommendation engine which can lead to lost sales.


In this article, we outline our experience with recommendation engines and  propose three different kind of algorithms that businesses can use and implement to their need. But what is a recommendation engine?  When a customer uses Amazon to buy a book, the site automatically proposes other related books that the user might like. These books somehow match the taste of the customer and thus are recommended.  So how does a recommendation engine work?  A recommendation engine consists of an algorithm that processes website users’ data and recommend products tailored to the taste of those users.

You might think recommendation engines are only for large corporations with big amounts of data. You might be right… however there are multiple forms of recommendation algorithms that can fit into any business. In general, we distinguish three main methods of various recommendation engines. Those are content based filtering, collaborative or item based filtering and hybrid engines.


The first method is content based filtering. This algorithm recommends products which are similar to the ones that a user has liked in the past. As an example, John visits a platform which he uses to stream movies. In this case, he likes “movie 1”. This information is collected in John’s profile for the use of this recommendation engine. Now that we know that John liked “movie 1”,  using a metric (such as cosine similarity, Euclidean distance) allows us to predict movies similar to “movie 1” based on information such as keywords, actors, director and the genre of movies.

The second method of recommendation engines are collaborative filters. This method can further be classified into two types: user-based filtering and item-based filtering. User-based filtering is a system that recommends products to a user that similar users have liked. For example, let’s say John and Jane have a similar interest in books. A new book appears on the market that John has read and liked. It is therefore, highly likely that Jane will like it too. Therefore, the system recommends this book to Jane.


The second type of collaborative filters is the Item-based filtering. This recommendation system is extremely similar to the content recommendation engine. It identifies similar items based on how people have rated it in the past. In this case, John, Jane and Alfred liked “book 1” and “book 2”. The recommendation system identifies “book1” and “book2” as similar items. Therefore, if Mark buys the “book 1”, the algorithm also recommends “book 2” to him.

The third and last method is the hybrid recommendation engine. This is a combination of the previous two recommendation algorithms. In essence, we combine data related to users and data related to the product in order to fit the recommendation as accurate as possible.

Recommendation engines can increase revenue and customer satisfaction for businesses.  Users can be provided with products or services they want to add to their basket or want to interact next with. DataCrunch builds recommendation engines to promote and increase product sales/services. Like to know more? Then contact us through this link.

DataCrunch V.O.F.