Let’s say all you know about your customers is their purchase history. With this data you want to generate individual recommendations representing cross-selling opportunities. How? Association Rule Learning.

In my opinion, this is a great approach if you find yourself limited by the lack of additional data that could give further context to a business transaction.

Even in cases where you do have additional data on which base your predictions, you might find yourself reverting back to using just transactions (perhaps as a supplement), simply because it often performs just as good or better than more complex recommendation systems.

The concept of Association Rules (also known as Market basket analysis) has been around a very long time. And for good reason. It’s simple to implement, easy to exlain, and delivers great practical results.

Code

I’ve written what’s essentially a wrapper for the arules library in R. The added functionality generates a dataframe with final predictions, rather than just the rules themselves.

Install the package directly from github with devtools. Run the first line if you do not currently have devtools installed.

# install.packages('devtools') 
devtools::install_github('HenrikVarmer/aruleRec')

aruleRec function

The aruleRec() function takes a dataframe as input in tidy (long) format with one observation (customer-item pair) per row. The input data.frame must contain two columns: a customer ID and a product ID. The data.frame and column names are provided as arguments to the function, along with the hyperparameters for rule mining - confidence, support, minlen, and maxlen.

The aruleRec() function returns a dataframe with customers and corresponding cross-selling recommendations and rule quality parameters. Only customers with recommendations are returned.

Example input data structure:

The ideal input data structure contains customers in one column, and products in another. Each row is one observation (i.e. one set of customer - item keys). Notice that one customer can have multiple rows, since a row represents one purchase of one item.

Example usage of aruleRec() function:

Load your data and apply the aruleRec() function. Supply the dataframe, the item column name, the customer column name, as well as parameters for the association rule mining algorithm. The optional keep_all argument defines weather all customers should be returned, or only those where the algorithm can make a recommendation.

dat <- read.csv("your_sales_data.csv")

recommendations <- aruleRec(data        = dat,      # dataframe
                            productkey  = Item,     # item ID column
                            customerkey = Customer, # contact ID column
                            keep_all    = FALSE,    # return all contacts or just rows with recommendations
                            minlen      = 2, 
                            maxlen      = 20, 
                            support     = 0.01, 
                            confidence  = 0.1)

head(recommendations, 3)

Example output:

In the output dataframe returned from the aruleRec() function, the ‘item_history’ column constitutes the customer purchase history. The ‘recommendation’ column indicates the cross-selling opportunity based on the mined association rules, along with key metrics support (the frequency with which the rule set appears in data), confidence (how often the rules is true), lift (how non-random the rules is), and count (how often the rule occurs in data).

This library relies on the arules package for mining association rules and simply provides a wrapper for easy generation of recommendations with tidy datasets.