2) Introduction to Non-Personalized Recommendations using Association Rules for Beginners

Hello! This article is a part of a full tutorial of Recommendation Systems which you can find here. I strongly recommend to follow the agenda of this tutorial if you want to master this topic.

What is Non-Personalized Recommendations

In previous article we already discuss the main difference between personalized and non-personalized systems but let’s explain it again.

Non-personalized recommendations are usually where the most sites start because it’s easy and doesn’t requires to know anything specific about the customers, because customers individual likes and dislikes aren’t used for customized suggestions and the same recommendation is given to all of them. That’s why its non-personalized.

For example: what if a completely new user visits an e-commerce site. How does the site go about recommending products to the user in this scenario? While we don’t have any past history of that user we can not create personalized offers but we can suggest him the most popular items in the store. For example: bestsellers in book store, top movies on Netflix account, most popular news articles on news feed.

Also, websites have many non-identified visitors. While they aren’t registered you don’t have any information about them to show personalized content and recommending the most popular content remains the only option in this case.

If item is liked by majority of population, it has a big chance to be liked by you

You might think, that non-personalized recommendations should be shown until the system knows enough about the user and then show personalized content but keep in mind that humans are flock animals by nature and most people want to know what content are the most popular.

Popularity is a good way to create non-personalized recommendation because if item is liked by majority of population, it has a big chance to be liked by other users as well.

Keep in mind that there are industries where popularity is not the right way to recommend something. For example music or clothing industry. Also, online stores have million of products and there maybe products that are less famous but some group of people may buy them. In this case popularity is not the solution, but personalized recommendation system solves the problem of discovery by recommending products according to user’s preferences. We will discuss personalized recommendation engines later, in the next articles.

Now we will discuss kind of non-personalized recommendations that all of us have seen on Amazon or other e-commerce web-sites.

E-Commerce Recommendations

Here are the common forms of non-personalized e-commerce recommendations that you have seen on your online shops web-site:

• Recommending items that can be add to your basket
• Recommending pairs of items/services that go well together
• Recommending products that are useful when buying some product

E-commerce Recommendations

Let’s imagine you are a sales assistant and someone buys the phone. You know, there is a high chance he will also buy a phone cover because these products are frequently bought together.

So it would be wiser for us to recommend phone covers to that user (cross-selling). In other words, we are recommending products that are associated to another product. Here the Association Rule concept comes into play.

Data Mining: Data mining is used to discover interesting and previously unknown patterns from datasets. Association rule mining is a popular and well researched method of data mining for discovering interesting relations between items in the databases.

Association Rule Mining: The goal of association rule mining is to know which items are frequently bought together. Given a set of transactions, find rules that will predict the occurrence of an item (Item B) based on the occurrence of other items (Item A) in the transaction. For example: Predict the chance of user buying a phone cover (Item B) if he already bought the phone (Item A) and if the chance is high enough then recommend phone cover to someone who are buying the phone.

Keep in mind that the implication means co-occurrence and not causality! These items are just frequently bought together but we can not assist that buying one item is cause of buying the other one.

Before we discuss all of the steps to generate association rules let’s see some more use cases and examples:

Association rule recommending the items that can be add on your basket

Now imagine you are a data scientist and you have a client who runs a retail store. Your client gives you all of the transactions data. This data consists of items bought in the store by customers over a period of time. You are going to use Market Basket Analysis (MBA) which uses Association Rule Mining on the given transaction data to help your client to boost sales. Market basket analysis is the process of finding the buying habits of people.

Here is the transactions data containing only 5 transactions

You can see transactions numbered 1 to 5 here and each transaction shows items bought in that transaction. You can tell to your client what items do customers frequently buy together by generating rules called Association Rules.

Before we go to exact steps and calculations let’s find association rule just by looking into the data. We can see that Diapers and Beer are frequently bought together because they appear 3 times in 5 transactions:

• {Diapers, Beer} is an example of a frequent Itemset because appear frequently together
• {Diapers} -> {Beer} is an example of an association rule which means that there is a high chance to buy beer when you are buying diapers
• Diapers and Beer is a very old and well-known example of association rules. The rule suggests that a strong relationship exists between the scale of diapers and beer because many customers who buy diapers also buy beer. Retailers can use this type of rules to help them identify new opportunities for cross-selling their products to customers.

If you don’t know who buys diapers and then beer try to guess :) and leave me the answer in the comments

Okay, we have some association rules based on our data and we know which items are frequently bought together. Let’s see what our client can do in terms of marketing strategies with this kind of information:

• Changing the store layout according to trends (If you put frequently bought items far from each other customer will spend more time in the store, or somethings maybe it’s better to put them together)
• Stock management (What other products should the store stocks up?)
• Cross sale offers (Buying this, let’s add this with the price of x)
• Customized emails with add-on sales
• Create a catalogue design and so on.

Next article is an introduction to association rules. Continue reading here