CSP 2017 Online Program

Sequential Pattern Mining in Real-Time Marketing with Backward Match Algorithm (303462)

Keywords: Real time marketing, sequential data mining, backward-match algorithm, retail credit card services

Traditional sequential data mining has been focusing on discovering a series of patterns shared by customers. In mortar-and-bricks environment, it has been utilized in studying customers buying behavior, products association etc. Nowadays, mobile and social platform provides retailers enormous opportunities to capture customer’s immediate transactions and promote in a real-time fashion. In this paper, we develop a novel backward-matching algorithm and hybrid it with CSPADE (Constraints Sequential Pattern Discovery using Equivalence class) algorithm, which enables us to customize product recommendations right after each transaction. In our study, there are three steps in customizing recommendation (1) using CSPADE to mine customer shopping patterns and generate a promotable sequences; (2) backward-matching customer’s immediate transactions with patterns in the promotable sequences; (3) prioritizing matched promotable product based on predictive power (confidence) or incrementaility (lift) and make recommendation. We demonstrate this novel algorithm on a credit card client in a case study. This hybrid technic can also be easily adapted into other fields, such as website advertisement, disease prevention and diagnose, fraud detection and etc.