Online Retail Pattern Quality Improvement: From Frequent Sequential Pattern to High-Utility Sequential Pattern

Abstract

There has been a change in people's shopping behavior, especially during the Covid-19 pandemic, from what traditionally requires direct face-to-face meetings between sellers and buyers, to virtual face-to-face through various shopping media. All activities carried out by customers, click streams performed, items purchased, the number of items including the price will be recorded in a log. Activity records in the log are very useful to be able to find out the pattern of activity sequences from customers, especially the order of items purchased by customers. However, the management certainly needs more knowledge, not just the order of goods that are often purchased by customers. Does the order of items purchased also provide maximum profit? There have been many methods to get frequent sequential patterns from customer activities, but getting a pattern that chooses more quality by adding utility value needs to be considered. In this research, the method used to obtain frequent sequential patterns is using PrefixSpan (Prefix-projected Sequential PAtterN) and the method used to obtain a high-utility sequential pattern is the USpan (Utility Sequential PAtterN) method. USpan is applied to the BMS (Blue Martini DataSet) dataset, which is the dataset used in KDD (Knowledge Discovery in Databases) CUP 2000 which consists of clickstream data from an e-commerce. The experimental results show that the frequent sequential pattern will always appear in the high-utility sequential pattern but not vice versa. It is certain that a high-utility pattern must be sequential, but a sequential pattern is not necessarily a high-utility sequential. From the results of the high-utility sequential pattern, it can be used as input to provide recommendations to customers to carry out the shopping process on items that can provide greater profits. The conclusion of the research conducted is that the high-utility sequential pattern mining can produce a higher quality pattern than just getting a frequent sequential pattern.