In this book, recommendation engines are discussed in Chapter 8, Neural Network Algorithms. Basket analysis is a simpler way of learning recommendations. In basket analysis, our data contains only the information regarding what items were bought together. It does not have any information about the user or whether the user enjoyed individual items. Note that it is much easier to get this data than it is to get ratings data.

For example, this kind of data is generated when we shop at Walmart, and no special technique is required to get the data. This data, when collected over a period of time, is called transnational data. When association rules analysis is applied to transnational data sets of the shopping carts being used in convenience stores, supermarkets, and fast-food chains, it is called market basket analysis. It measures the conditional probability of buying a set of items together, which helps to answer the following questions:

• What is the optimal placement of items on the shelf? 
• How should the items appear in the marketing catalog?
• What should be recommended, based on a user's buying patterns?

As market basket analysis can estimate how items are related to each other, it is often used for mass-market retail, such as supermarkets, convenience stores, drug stores, and fast-food chains. The advantage of market basket analysis is that the results are almost self-explanatory, which means that they are easily understood by the business users.

Let's look at a typical superstore. All the unique items that are available in the store can be represented by a set,  = {item ₁ , item ₂ , . . . , item _m }. So, if that superstore is selling 500 distinct items, then  will be a set of size 500.

People will buy items from this store. Each time someone buys an item and pays at the counter, it is added to a set of the items in a particular transaction, called an itemset. In a given period of time, the transactions are grouped together in a set represented by , where   = {t ₁ ,t ₂ , . . . ,t _n }.

Let's look at the following simple transaction data consisting of only four transactions. These transactions are summarized in the following table:

Let's look at this example in more detail:

 = {bat , wickets, pads, helmet, ball }, which represents all the unique items available at the store.

Let's consider one of the transactions, t3, from . Note that items bought in t3 can be represented in the itemset_t3= {helmet,ball}, which indicates that a customer purchased two items. As there are two items in this itemset, the size of itemset_t5 is said to be two.