No-one could have predicted COVID-19 and its impact on our lives caught many of us by surprise. Businesses in different industries have been affected by the global pandemic to different extents, with hospitality, travel and retail taking the hardest hit. While it is unclear when all businesses across the UK will be back to operating in full swing, minimizing further losses is the priority for the majority of companies.
Hence, it is not surprising that the use of predictive analytics has become more popular than ever. Predicting and maximizing sales, minimizing losses and estimating customer attrition is what, hopefully, will help as many businesses as possible to stay afloat during these challenging times.
As appealing as it seems, for many Machine Learning sounds complicated and alien. When something is so theoretical and complex, it is difficult to know how it can be practically applied to your business. This is especially the case for companies that don’t have onsite developers and specialists.
The aim of this series is to outline and explain mainstream business applications of Machine Learning. In the present article we will provide a high-level overview of what Recommendation Engines are and how they are used.
Recommendation Engines i.e. “How did it know?”
What are they?
That Monday afternoon when you are about to buy Alexa Dot and you see…
-You may also like “Smart Alexa Light Bulb” … It will be voice activated and will work with Alexa...
So brilliant and simple. Yes, Amazon, I would love to control lights with my voice. While this may not be relatable to all, each one of us internet-users encounters work of Recommendation Engines daily.
Recommendation Engines are a class of Machine Learning techniques that use data filtering models to predict user preference. This can take many forms, such as product recommendation on an e-commerce websites, content recommendations on the personalized newsfeed and entertainment recommendations in media and streaming services.
How do they work?
Firstly, the underlying model is chosen for the algorithm. This will determine the type of data that will be collected and the mathematical methods that will be used by Machine Learning pipelines to calculate the predictions. There is a wide range of techniques available, but the most common methods implement collaborative filtering and content-based filtering models.
Collaborative filtering: This technique can filter out items that a user might like based on the reaction of similar users. It works by filtering through a large dataset of customers and finding smaller subsets that are similar to your customer. The items liked by the user and a similar group of users is combined to make a more extensive recommendation list.
Content-based filtering: This model operates by recommending similar items to the ones the user liked before. It filters through the dataset of products to find similar items to generate larger list of suggestions. This is largely limited by the user's previous shopping history.
Behind the scenes similarity between different products is calculated using mathematical algorithms such as cosine similarity and Euclidean distances, and statistical models such as Pearson’s coefficient.
While there are differences between the models, both of them form basis of a Machine Learning pipeline that works in a conceptually similar manner, where data is collected (implicitly/explicitly), data is stored, analyzed and filtered. Filtering is applied to data to generate recommendations that are presented to customers.
Where can you see them?
Hopefully, I didn’t put you to sleep. For me, it is very exciting to see how recommendation engines are applied across different industries. Recommendation engines are everywhere these days. In fact, some of the biggest brands engage with them daily. The list includes Netflix, Amazon, Spotify, Google and Googlereads… the list goes on.
Netflix uses a recommendation engine to present viewers with movie and show suggestions. Amazon, on the other hand, uses a recommendation engine to present customers with product recommendations. While each uses one for slightly different purposes, both have the same goal: to drive sales, boost engagement and retention, and deliver more personalized customer experiences.
These complicated sounding models have become very efficient over time and have helped many businesses to generate additional revenue from new and existent customers. Interestingly, according to Forbes – over 35% of purchases on Amazon are generated through product recommendations. Recommendation engine are everywhere and can be applied almost in every sector in different capacities.
Personally, I love it when Netflix and Spotify generate “top picks” for ME. I find it comforting to feel like Netflix knows me, however, I still struggle to decide what to watch (there is just too much choice).
How can you start?
Implementing recommendation engines may be the right choice for your business. If it is, next step would be to start designing the solution and implementing it. A great place to start is Microsoft Cognitive Services, which offers a range of products, such as Machine Learning tools that can be used by general customers.
However, if you want additional support and would like to receive help from professionals with years of experience in predictive analytics, partnering with a company that specializes in predictive analytics (such as Coeo) could be a great place to start.
Find out more
- Industries Most and Least Impacted by COVID-19 from a Probability of Default Perspective – September 2020 Update | S&P Global Market Intelligence (spglobal.com)
- COVID‐19 Pandemic in the New Era of Big Data Analytics: Methodological Innovations and Future Research Directions - Sheng - - British Journal of Management - Wiley Online Library
- Recommendation Engines - ScienceDirect
- Build a Recommendation Engine With Collaborative Filtering – Real Python
- Essentials of recommendation engines: content-based and collaborative filtering | by Jonathan Leban | Towards Data Science
- Recommendation Engine - an overview | ScienceDirect Topics