What are Recommender Systems and Why Should I Care? What are Recommender Systems and Why Should I Care?
Introduction Every time you want to acquire a new product or service from a physical store or business, you may go there and ask... What are Recommender Systems and Why Should I Care?

Introduction

Every time you want to acquire a new product or service from a physical store or business, you may go there and ask to speak to a salesperson. You will tell them what exactly it is you want from said product, such as a laptop specifically with an SSD and a touchscreen, and they, in turn, will ask you questions to better understand what it is you want. From there, they’ll make some recommendations based off of your conversation. Sometimes you rely on this discourse to decide on what product is right for you.

That job of a salesperson is what a recommender system does but in the digital world. Whenever you search for a product or service it will provide advice based on your past purchases and searches, and based on the purchases that customers who bought similar items as you and what they bought after those items. A recommender system will try to find the best products or services according to your needs.

In the case of a recommender system, instead of asking you about your needs, it will infer them based on your searches, your purchases, and the data available about you and your behavior.

What exactly is a recommender system?

Is it the algorithm? Is it the model? Is it the result of the model? Is it a text saying what to recommend next? Is it a set of lines?

The answer is that a recommender system is not a single thing but the sum of multiple components. That’s why the word system is in the term. When you read:

This [product / destination / service / friend / artist] might interest you.

The previous phrase is the result of a process. Data collection, data preprocessing, data modeling, feature engineering, the application to some machine learning algorithms based on the input data, model selection, model evaluation, and then the delivery of the results. This process considers data storage, data processing, data engineering, and data science.

A recommender system is a desirable data product but is useless if you can’t act based on its results. Is advisable to think, before modeling, what you can do once you have the results.

A recommendation will be customized, based on historical data from users that made similar purchases. If we can keep track of the data we can predict what is coming next.

Association rules are a simple and effective approach to make recommendations to your customers based their transactions. It can identify repeated transactions over time and spot patterns from the data. A rule will have the form of “X% of the customers who buy A also buys B”.

Now, we can recommend the item B to those customers who got A but have not bought B (yet).

How to use recommendation systems?

In Business

Many companies, before analyzing isolated datasets from multiple business processes, collect them all and put them together. Many analyses can be derived then from this particular dataset, such as the customers’ complete history, when they joined your business if they were referred, and if so, by whom?

Having the history of your customer’s transactions together is a great start to mining relevant insights. You can find hidden patterns in their behavior or get evidence to validate any hypotheses that have been in the business for a long time.

Next Best Offer (NBO)

The NBO consists in analyzing past transactions of your customers in order to find relevant combinations of products. For example, you may find that many customers get product A and product B, so you can take actions recommending product B to those customers who have acquired product A in the past.

You can put all the transactions in a bag and analyze them or you can analyze them including the time dimension, by sequence.

If you use all transactions you will get more results than using a time-wise analysis, but both approaches are worth doing. My recommendation is to start with all transactions, and then explore in detail what are the relations between products considering its sequence.

Next Best Action (NBA)

NBA is similar to NBO but is related to taking proactive actions regarding communications or contacts with customers. From the transactions log, you can find out strategies of how to engage and increase the loyalty of groups of customers.

The NBA will tell you when is best to contact them, what type of contact method should be used, and through which channel of communication. You can schedule and automatize these actions in your systems.

Next Best Product

This is more known by you if you buy things on Amazon, watch movies on Netflix, listen to music online, etc. Often these companies make suggestions for you, recommending things you might like based on the study of the transaction of customers, you included, and then matched with your interests.

Conclusions

A recommender system is something you implement using the data you already have on your customers. It will help you to increase revenues and optimize resources, while increasing customer loyalty by knowing them better and understanding their needs.

 

Diego Arenas

Diego Arenas, ODSC

I've worked in BI, DWH, and Data Mining. MSc in Data Science. Experience in multiple BI and Data Science tools always thinking how to solve information needs and add value to organisations from the data available. Experience with Business Objects, Pentaho, Informatica Power Center, SSAS, SSIS, SSRS, MS SQL Server from 2000 to 2017, and other DBMS, Tableau, Hadoop, Python, R, SQL. Predicting modelling. My interest are in Information Systems, Data Modeling, Predictive and Descriptive Analysis, Machine Learning, Data Visualization, Open Data. Specialties: Data modeling, data warehousing, data mining, performance management, business intelligence.

Open Data Science - Your News Source for AI, Machine Learning & more