Using Machine Learning and Azure Search to improve conversion rates through recommendations
Recommendations are a really powerful tool to expose your users to more content in your catalog. It can be a technique to help increase sales, and also a way to assist user navigation of you web or mobile application. You probably have seen recommendations in some form, such as an e-commerce retail site where they showed you items frequently bought together, along with the one you’re currently looking at. Or maybe in a movie streaming site where it’s suggested, “Users who liked this movie, also liked this.” This technique is commonly called Item Based Recommendations.
Another type of recommendations you might have seen is called User Based Recommendations or collaborative filtering, which is the idea of providing user specific recommendations based on what is known about the user and the similarities they have to other users.
The interesting thing about recommendations is that it doesn’t apply to just ecommerce scenarios. It can apply to virtually any type of search application. For example, if you have a knowledge base of articles, you might want to suggest alternate articles based on what other users are reading or if you had a support forum where users could search for answers to questions, you might want to recommend other alternative answers based on other user’s behavior. In the case of Nubimetrics, recommendations are used to allow marketers to make decisions that enhance their sales based on over 30 million pieces of competitive information across 16 countries:
Many e-commerce marketplaces have millions of items for sale, where being seen by potential customers can be difficult. The easy way to climb up the ladder and be listed among the first results in the list is to pay for more exposure. But, will paying more result in more visits, and therefore potentially more sales? That’s where Nubimetrics comes in: based on all the information we gather from the marketplace on a daily basis we can recommend the appropriate level of exposure for each item the Seller publishes. As the exact position in the list is based on how many people are paying and how much you climb up depends on your original position, we can help the seller boost sales while saving them money. – Guillermo J. Bellmann, Cloud Architect, Nubimetrics
Recommendations can be a bit intimidating, so in this tutorial, I will focus on the core essentials of what is needed to get a solid implementation of recommendations working with an Azure Search index. I will also focus on Item Based Recommendations, but the same technique can be used for User Based Recommendations. From there, you can start building on it to further improve the recommendations, without having to be a Data Scientist.
You can get started with this tutorial by visiting GitHub where you can also see a live demonstration of what you will build. I hope you find this helpful and if you have any questions or feedback on this, please let me in the comments below.
Source: Microsoft Azure News