Case Background
When we are shopping online or when we receive an advertisement email from a shopping website that is recommended based on personalized products, how does the shopping website calculate and summarize such information? In fact, the whole process is called shopping basket analysis (also known as association analysis) or recommendation engine. The engine at the bottom layer collects the information related to the shopping habits or preferences of each person, and then discover the hidden rules and commercial knowledge by using the analysis tool.
If people buy pasta and wine today, they are usually interested in the pasta sauce. Once this rule is found, you will be recommended with some pasta sauce the next time when you buy pasta and wine. The above description will generate a set of association rule:
IF {pasta, wine} THEN pasta sauce
The Apriori algorithm is the most famous and widely applied method to generate this rule. The first part of the rule is called "Antecedent", while the latter part is called "Consequent". The relevant pointer is also used to confirm the reliability of each rule, such as the Antecedent Support, Rule Support%, Rule Confidence%, and Rule Lift.
To embrace the era of Big Data, a hypermarket that meets the characteristics of 3V (Volume, Variety, and Velocity), starts to conduct in-depth analysis of big data. The analysists analyze the internal transaction data and member’s data generated on a daily basis to find the best product recommendations and arrangements.
This case is divided into two parts. First, we establish the association rules of the products based on the existing data, which are then deployed. When new transaction data comes in the future, these rules can be applied to generate the latest and suitable recommended items for the customer in real time.
Solution
1. Association Rule Analysis
After importing the transaction data of the products in the existing database and the company's product related information (name and price), it may conduct the association analysis to generate many association rules, each of which will display the Antecedent and Consequent, as well as the related indicators calculated, such as the Antecedent Support, Rule Support, Rule Confidence, and Rule Lift.
According to the association rules, we can give the following interpretation, if a customer usually (or usually supports) to buy wine, pasta and garlic (Antecedent), they will also buy pasta sauce (Consequent). With this rule confidence, we believe this rule is useful.
2. Deployment of Association Rules
The recommendation engine is to help recommend the goods that will be possibly purchased based on the member's preferences. Moreover, it reduces the annoying advertising letters for members, achieving a win-win outcome.
Relevant Cases:
